Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Azriel RosenfeldSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works399 from public data
- 1,270
- 780
- 744
- Automatic text detection and tracking in digital video664
This work presents algorithms for detecting and tracking text in digital video that implements a scale-space feature extractor that feeds an artificial neural processor to detect text blocks.
- Towards Optimal Structured CNN Pruning via Generative Adversarial Learning585
This paper proposes an effective structured pruning approach that jointly prunes filters as well as other structures in an end-to-end manner and effectively solves the optimization problem by generative adversarial learning (GAL), which learns a sparse soft mask in a label-free and an end to end manner.
- 495
- Camera-based analysis of text and documents: a survey487
A survey of application domains, technical challenges, and solutions for the analysis of documents captured by digital cameras, and some sample applications under development and feasible ideas for future development is presented.
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks366
This paper proposes a trellis encoder-decoder network (TEDnet) for crowd counting that achieves the best overall performance in terms of both density map quality and counting accuracy, and proposes a new combinatorial loss to enforce similarities in local coherence and spatial correlation between maps.
- The Indexing and Retrieval of Document Images: A Survey322
A survey of methods developed by researchers to access and manipulate document images without the need for complete and accurate conversion is provided.
- Video summarization by curve simplification319
A simple video player is described that displays the keyframes sequentially and lets the user change the summarization level on the fly with a slider, and an approach to automatically selecting a summarizationlevel that provides a concise and representative set of keyframes is described.
- 316
- 294
- 234
- 233
- Review of Classifier Combination Methods224
This chapter introduces different categories of classifier combinations and introduces a retraining effect and effects of locality based training as important properties of classifiers combinations.
- 215
- 199
- 197
- 193
- Handbook of Document Image Processing and Recognition191
The Handbook of Document Image Processing and Recognition is a comprehensive resource on the latest methods and techniques in document image processing and recognition that enables the reader to make an informed decision for their specific problems.
- 174
- Simultaneous estimation of image quality and distortion via multi-task convolutional neural networks172
- 160
- Future of software development with generative AI159
The software development industry needs new tools to understand the potential, limitations, and risks of generative AI, as well as guidelines for using it, and proposes four primary scenarios, model trajectories for transitions between them, and reflect against relevant software development operations.
- 157
- 149
- Exploiting Kernel Sparsity and Entropy for Interpretable CNN Compression142
KSE is capable of simultaneously compressing each layer in an efficient way, which is significantly faster compared to previous data-driven feature map pruning methods, and significantly outperforms state-of-the-art methods.
- 136
- 136
- Recovery of temporal information from static images of handwriting126
It is demonstrated how many seemingly ambiguous situations can be resolved by the derived clues and the knowledge of the writing process, and several examples to illustrate the approach.
- 124
- Preserving Privacy in Federated Learning with Ensemble Cross-Domain Knowledge Distillation123
This work develops a privacy preserving and communication efficient method in a FL framework with one-shot offline knowledge distillation using unlabeled, cross-domain, non-sensitive public data and shows that it outperforms baseline FL algorithms with superior performance in both accuracy and data privacy preservation.
- Compressed-domain video indexing techniques using DCT and motion vector information in MPEG video115
A frame-type independent representation of the various types of frames present in an MPEG video in which al frames can be considered equivalent is developed and enable fast archiving, indexing, and retrieval of video.
- Classification of document pages using structure-based features112
The approach to classification is based on “visual similarity” of layout structure and is implemented by building a supervised classifier, given examples of each class, using decision tree classifiers and self-organizing maps.
- Multiscale segmentation of unstructured document pages using soft decision integration109
An algorithm for layout-independent document page segmentation based on document texture using multiscale feature vectors and fuzzy local decision information is presented.
- Text enhancement in digital video using multiple frame integration108
Experiments conducted on several video sequences show that the multiple frame based enhancement scheme can improve the accuracy of commercial off-the-shelf OCR considerably.
- Structural similarity for document image classification and retrieval107
A novel approach to defining document image structural similarity for the applications of classification and retrieval by encoding each document and model the spatial relationships between them by recursively partitioning the image and computing histograms of codewords in each partition.
- NAS-Count: Counting-by-Density with Neural Architecture Search105
This work automates the design of counting models with Neural Architecture Search (NAS) and introduces an end-to-end searched encoder-decoder architecture, Automatic Multi-Scale Network (AMSNet), utilizing a counting-specific two-level search space.
- Projection Convolutional Neural Networks for 1-bit CNNs via Discrete Back Propagation100
This paper introduces projection convolutional neural networks with a discrete back propagation via projection (DBPP) to improve the performance of binarized neural networks (BNNs) and learns a set of diverse quantized kernels that compress the full-precision kernels in a more efficient way than those proposed previously.
- 99
- 98
- 97
- Archiving, indexing, and retrieval of video in the compressed domain95
This paper describes feature extraction and key frame indexing and retrieval techniques that are directly applicable to compressed video that enable fast parsing and archiving of video.
- 93
- 92
- 91
- 89
- Multi-UAV Mobile Edge Computing and Path Planning Platform Based on Reinforcement Learning87
This work introduces a new multi-UAV Mobile Edge Computing platform, which aims to provide better Quality-of-Service and path planning based on reinforcement learning to address issues of UAV use in complex environments with obstacles and cooperation between UAVs.
- 87
- Video retrieval using spatio-temporal descriptors87
A novel methodology for implementing video search functions such as retrieval of near-duplicate videos and recognition of actions in surveillance video, with the distinct advantage that there is no need for detection or recognition of body parts.
- 84
- Circulant Binary Convolutional Networks: Enhancing the Performance of 1-Bit DCNNs With Circulant Back Propagation84
Newcirculant filters (CiFs) and a circulant binary convolution (CBConv) are proposed to enhance the capacity of binarized convolutional features via the authors' circulants back propagation (CBP).
- 83
- Applying algebraic and differential invariants for logo recognition81
A novel application of algebraic and differential invariants to the problem of logo recognition is presented and shape descriptors for matching that are unique and independent of the point of view are defined.
- 76
- Cross-Language Entity Linking75
A new test collection is created to evaluate cross-language entity linking performance in twenty-one languages and presents experiments that examine issues such as: the importance of transliteration; the utility of cross- language information retrieval; and, the potential benefit of multilingual named entity recognition.
- Identifying sports videos using replay, text, and camera motion features72
The extraction of features that enable identification of sports videos directly from the compressed domain of MPEG video, including detecting the presence of action replays, determining the amount of scene text in vide, and calculating various statistics on camera and/or object motion are discussed.
- 71
- Building an information retrieval test collection for spontaneous conversational speech71
The use of search-guided relevance assessment is described to create a test collection for retrieval of spontaneous conversational speech and baseline results for ranked retrieval are presented to illustrate use.
- Handwritten Arabic text line segmentation using affinity propagation70
A novel graph-based method for extracting handwritten text lines in monochromatic Arabic document images using primary components which define the line and assignment of diacritic components which are more difficult to associate with a given line is presented.
- A New Algorithm for Detecting Text Line in Handwritten Documents70
This work model text line detection as an image segmentation problem by enhancing text line structure using a Gaussian window, and adopting the level set method to evolve text line boundaries.
- 70
- 68
- 68
- 67
- Text and non-text separation in offline document images: a survey67
The technical challenges of performing text/non-text separation are summarized, and offline document images are categorized into different classes according to the nature of the challenges one faces, in an attempt to provide insight into various techniques presented in the literature.
- 67
- 67
- VideoTrails66
A novel technique is described which reduces a sequence of MPEG encoded video frames to a trail of points in a low dimensional space and tracks the interaction of clusters over time, laying the groundwork for the complete analysis and representation of the video’s physical and semantic structure.
- 65
- Scalable Coverage Path Planning of Multi-Robot Teams for Monitoring Non-Convex Areas64
The results show that SCoPP is superior in terms of mission completion time; its computing time is found to be under 2 mins for a large map covered by a 150-robot team, thereby demonstrating its computationally scalability.
- 63
- 62
- 61
- Smart mask – Wearable IoT solution for improved protection and personal health60
The Smart Mask is envisioned, an IoT supported platform and ecosystem aiming to prevent and control the spreading of COVID-19 and other respiratory viruses and the integration of sensing, materials, AI, wireless, IoT, and software will help the gathering of health data and health-related event detection in real time from the user as well as from their environment.
- Adaptive Hindi OCR using generalized Hausdorff image comparison60
An adaptive Hindi OCR system was applied to a complete Hindi--English bilingual dictionary and a set of ideal images extracted from Hindi documents in PDF format and results show the recognition accuracy can reach 88% for noisy images and 95% for ideal images.
- A Dataset for Quality Assessment of Camera Captured Document Images59
A dataset of camera captured document images containing varying levels of focal-blur introduced manually during capture is presented and three recent methods for predicting the OCR quality of images on this dataset are presented.
- FORGERY DETECTION BY LOCAL CORRESPONDENCE58
This thesis approaches the off-line problem by establishing a local correspondence between a model and a questioned signature, and performs skilled forgery detection by examining the writer-dependent information embedded at the sub-stroke level and trying to capture unballistic motion and tremor information in each stroke segment, rather than as global statistics.
- 54
- Page Segmentation and Zone Classification: The State of the Art54
The purpose of this review is to analyze progress made in page segmentation and zone classification and suggest what needs to be done to advance the field.
- GiB : A ${G}$ ame Theory ${I}$ nspired ${B}$ inarization Technique for Degraded Document Images53
The experimental results show that GiB (Game theory Inspired Binarization) outperforms competing state-of-the-art methods in most cases.
- 52
- 52
- Federated Learning With Privacy-Preserving Ensemble Attention Distillation51
This work proposes a privacy-preserving FL framework leveraging unlabeled public data for one-way offline knowledge distillation in this work and achieves very competitive performance with more robust privacy preservation based on extensive experiments on image classification, segmentation, and reconstruction tasks.
- 50
- 50
- 49
- Document Image Retrieval Based on Layout Structural Similarity.49
An approach to computing a geometrically invariant structural similarity, and use this measure to search document image databases, and results demonstrating effectiveness and efficiency of structure queries with respect to human relevance judgments are presented.
- 49
- 49
- 48
- An introduction to vectorization and segmentation48
Some of the basic issues and trends in the vectorization and segmentation of graphics are overviewed, and a short list of relevant references on these topics are provided.
- 47
- 46
- 46
- 45
- The Segmentation and Identification of Handwriting in Noisy Document Images45
This paper presents an approach to the problem of segmenting and identifying handwritten annotations in noisy document images, and finds that the reliable results can be achieved at the word level with a classification accuracy of 97.3%.
- Towards Generic Image Manipulation Detection with Weakly-Supervised Self-Consistency Learning44
Weakly-supervised image manipulation detection, such that only binary image-level labels (authentic or tampered with) are required for training purpose, and two consistency properties are learned: multi-source consistency (MSC) and inter-patch consistency (IPC).
- 44
- 43
- Text enhancement in digital video43
Experiments show the enhancement approach improves OCR accuracy considerably, and consists of two main procedures, resolution enhancement based on Shannon interpolation and text separation from complex image background.
- 42
- A robust stamp detection framework on degraded documents42
This paper introduced effective algorithms to address the problem that stamps often spatially overlay their background contents, which give the approach significant advantages in detection accuracy and computation complexity over traditional Hough transform method in locating candidate ellipse regions.
- 42
- 40
- 39
- Building mosaics from video using MPEG motion vectors39
A novel way of creating mosaics from an MPEG video sequence is presented that explicitly compute camera motion between frames and deduce the camera motion directly from the motion vectors encoded in the MPEG video stream.
- 38
- Logical Labeling of Document Images Using Layout Graph Matching with Adaptive Learning38
This system is able to learn a model for a document class, use this model to label document images through graph matching, and adaptively improve the model with error feed back.
- 38
- Word level script identification for scanned document images37
Experimental results show the capability of Gabor filter to capture script features and the effectiveness of these three classifiers for script identification at the word level.
- Language identification for handwritten document images using a shape codebook36
This paper proposes a novel approach to language identification for documents containing mixture of handwritten and machine printed text using image descriptors constructed from a codebook of shape features, which is easily extensible and does not require skew correction, scale normalization, or segmentation.
- Classification of document page images based on visual similarity of layout structures36
This work classified UW-I document images based on 'visual similarity' of the layout structure by building a supervised classifier, given examples of the class, using the OC1, a decision tree classifier.
- 35
- 35
- Special-effect edit detection using VideoTrails: a comparison with existing techniques35
The performance of the Video Trails-based algorithms, with other existing special effect edit-detection algorithms within the literature are discussed.
- Deformable Gabor Feature Networks for Biomedical Image Classification34
A deformable Gabor convolution (DGConv) is introduced to expand deep networks interpretability and enable complex spatial variations and DGFN is introduced for addressing deep multi-instance multi-label classification on the INbreast dataset for mammograms and on the ChestX-ray14 dataset for pulmonary x-ray images.
- Anti-bandit Neural Architecture Search for Model Defense34
This paper defends against adversarial attacks using neural architecture search (NAS) which is based on a comprehensive search of denoising blocks, weight-free operations, Gabor filters and convolutions, and incorporates a new operation evaluation measure and search process based on the lower and upper confidence bounds (LCB and UCB).
- 34
- Leaf-Based Plant Disease Detection and Explainable AI33
This research survey provides a comprehensive understanding of common plant leaf diseases, evaluates traditional and deep learning techniques for disease detection, and summarizes available datasets, and explores Explainable AI (XAI) to enhance the interpretability of deep learning models' decisions for end-users.
- 33
- MIDPhyNet: Memorized infusion of decomposed physics in neural networks to model dynamic systems33
The results clearly demonstrate the performance benefits of the hybrid architecture over both purely data-driven models and state-of-art hybrid models in terms of generalizability and extrapolability.
- Machine-assisted authentication of paper currency: an experiment on Indian banknotes33
A system developed for discriminating fake notes from genuine ones and apply it to Indian banknotes is described and the ability of the embedded security aspects is thoroughly analysed for detecting fake currencies.
- 33
- 33
- 33
- 32
- THE DEVELOPMENT OF A GENERAL FRAMEWORK FOR INTELLIGENT DOCUMENT IMAGE RETRIEVAL32
The general framework, feature extraction modules, query capabilities, a graphical query interface, and the application interface are introduced and each component of the system is demonstrated and how the query mechanisms can be used to handle both content and structural queries eeectively.
- 31
- 31
- 31
- 31
- 30
- Binarized Neural Architecture Search29
This work introduces channel sampling and operation space reduction into a differentiable NAS to significantly reduce the cost of searching and achieves a performance comparable to NAS on both CIFAR and ImageNet databases.
- 29
- 29
- 28
- Mobile Retriever: access to digital documents from their physical source28
An image based document retrieval system which runs on camera enabled mobile devices that uses token triplets that define the orientation of three corresponding tokens to effectively prune the false positives and identify the correct page to retrieve.
- 28
- An appearance-based approach for consistent labeling of humans and objects in video28
An approach for consistently labeling people and for detecting human–object interactions using mono-camera surveillance video based on a robust appearance-based correlogram model combined with histogram information to model color distributions of people and objects in the scene is presented.
- 28
- 28
- Long term 5G network traffic forecasting via modeling non-stationarity with deep learning27
A deep learning model is developed that incorporates stationary processes into a well-designed hierarchical structure and models non-stationary time series with multi-scale stable features, showing potential to address broader engineering problems in 5G network traffic forecasting.
- Shape codebook based handwritten and machine printed text zone extraction27
A novel method for extracting handwritten and printed text zones from noisy document images with mixed content using Triple-Adjacent-Segment (TAS) based features which encode local shape characteristics of text in a consistent manner is presented.
- 27
- Defending against Data-Free Model Extraction by Distributionally Robust Defensive Training26
This work proposes a Memory and Computation efficient defense approach, named MeCo, to prevent DFME from happening while maintaining the model utility simultaneously by distributionally robust defensive training on the target victim model.
- Learning modulation filter networks for weak signal detection in noise26
A weak signal dataset is established that contains unmanned aerial vehicle (UAV) communication signals in a real-terrain environment and LMFNs, a two-stage optimization scheme which addresses filter learning, attention mechanism and classification in a unified framework are proposed.
- 25
- 25
- Developing high-level representations of video clips using VideoTrails25
A novel technique is described which reduces a sequence of MPEG encoded video frames to a trail of points in a low dimensional space and lays the groundwork for the complete analysis and representation of the video's physical and semantic structure.
- 24
- 24
- 24
- Exploring the Knowledge Transferred by Response-Based Teacher-Student Distillation23
This paper explores the transferred knowledge from a novel perspective: comparing the knowledge transferred through different teachers, and proposes an embarrassingly simple distillation framework named Efficient Distillation, which is effective on ImageNet with different teacher-student pairs.
- 23
- Binarized Neural Architecture Search for Efficient Object Recognition23
Binarized neural architecture search (BNAS), with a search space of binarized convolutions, is introduced to produce extremely compressed models to reduce huge computational cost on embedded devices for edge computing.
- 23
- TREC-10 Experiments at University of Maryland CLIR and Video.23
The University of Maryland Researchers participated in both the Arabic-English Cross Language Information Retrieval (CLIR) and Video tracks of TREC-10 to explore effective monolingual Arabic IR techniques and effective query translation from English to Arabic for cross language IR.
- Quantitative measurement of the performance of raster-to-vector conversion algorithms23
The performance of three approaches to raster-to-vector conversion — thinning, medial line finding, and line fitting — are compared using a methodology for measuring the performance of application-specific raster-to-vector conversion algorithms.
- Language-guided Human Motion Synthesis with Atomic Actions22
This paper decomposes complex human motions into a set of atomic actions, and employs a curriculum learning strategy to learn atomic action composition, which mitigates the overfitting problem commonly encountered in previous methods while enforcing the model to learn better motion representations.
- 22
- 22
- 22
- 22
- 22
- Document image understanding: integrating recovery and interpretation22
A "stroke platform" representation is provided which establishes a verifiable "link to the pixels" and demonstrate its usefulness for recovery tasks and introduces the concept of recovery into the document domain.
- Document Image Quality Assessment: A Survey21
This article surveys research on Document Image Quality Assessment (DIQA) and thoroughly review two standard measures to characterize document image quality: Optical Character Recognition-based and objective human perception-based.
- 21
- Cross-Language Entity Linking in Maryland during a Hurricane.21
The work on the cross-lingual task was significantly more involved, requiring development of robust, multiphased transliteration software, use of techniques in cross-language information retrieval, and reliance on a Chinese-toEnglish statistical machine translation system.
- Video retrieval of near-duplicates using κ-nearest neighbor retrieval of spatio-temporal descriptors21
This paper describes a novel methodology for implementing video search functions such as retrieval of near-duplicate videos and recognition of actions in surveillance video, using a threshold-free hierarchical space-time segmentation technique.
- Application of Structural and Topological Features to Recognize Online Handwritten Bangla Characters20
Two feature extraction methods are presented here that describe the transition from background to foreground pixels and vice versa and a combination of topological features and centre-of-gravity- based circular features where global information, local information, and Circular Quadrant Mass Distribution information have been extracted.
- 20
- Mosaicing of camera-captured document images20
This paper decomposes the complexity of solving the 8-dof transformation between image pairs into two problems, that is, rectification and registration, and presents a method for composing document mosaics from camera-captured images.
- 20
- SOAR: Scene-debiasing Open-set Action Recognition19
A novel method, called Scene-debiasing Open-set Action Recognition (SOAR), which features an adversarial scene reconstruction module and an adaptive adversarialscene classification module that prevents the decoder from reconstructing the video background given video features, and thus helps reduce the background information in feature learning.
- Scene Text Detection via Integrated Discrimination of Component Appearance and Consensus19
Experiments show that the proposed approach to scene text detection reaches the state of the art in both detection accuracy and computational efficiency, and in particularly, it performs best when dealing with low-resolution text in clutter backgrounds.
- 19
- Robust Point Matching for Non-Rigid Shapes: A Relaxation Labeling Based Approach19
Experiments demonstrate the effectiveness of the approach: it outperforms shape context and TPS-RPM algorithms under non-rigid deformation and noise on a public data set.
- 19
- 18
- Signature-Based Document Image Retrieval18
This work developed a fully automatic signature-based document image retrieval system that handles automatic detection and segmentation of signatures from document images and Translation, scale, and rotation invariant signature matching for document image retrieved.
- 18
- The TREC2001 Video Track: Information Retrieval on Digital Video Information18
There are now a variety of approaches available for searching and browsing through digital video archives, that do work, are scalable to larger archives and can yield useful retrieval performance for users, which has important implications in making digital libraries of video information attainable.
- 18
- A physics-aware learning architecture with input transfer networks for predictive modeling17
The goal of the OPTMA architecture is to facilitate greater exploitation of input space correlations between partial and full physics where they exist, by using a transfer neural network to transform the original inputs into modified inputs or latent features, where the partial physics operates on these artificially transformed features to produce the final prediction.
- CP-NAS: Child-Parent Neural Architecture Search for 1-bit CNNs17
A Child-Parent model is introduced to a differentiable NAS to search the binarized architecture under the supervision of a full-precision model (Parent) and achieves a comparable accuracy with traditional NAS on both the CIFAR and ImageNet databases.
- A camera-based mobile data channel17
A camera channel model is built to measure color degradation using information theory and it is shown that the capacity of the camera channel can be improved with the optimized color selection through color calibration, and a transmission bit rate is achieved that is faster than the average GPRS bit rate.
- 17
- Towards Compact 1-bit CNNs via Bayesian Learning16
A novel Bayesian Optimized compact 1- bit CNNs (BONNs) model is proposed, which has the advantage of Bayesian learning, to improve the performance of 1-bit CNNs significantly and introduces a new Bayesianlearning-based pruning method for 1- Bit CNNs, which significantly increases the model efficiency with very competitive performance.
- IOD-CNN: Integrating object detection networks for event recognition16
This work presents a novel unified deep CNN architecture which integrates architecturally different, yet semantically-related object detection networks to enhance the performance of the event recognition task.
- Font identification using the grating cell texture operator16
Experimental results show that the grating cell operator performs better than the isotropic Gabor filter, and the BPNN classifier can provide more accurate classification results than the WED classifier.
- Bootstrapping structured page segmentation16
This paper presents an approach to the bootstrap learning of a page segmentation model that has been applied to many structured documents such as dictionaries, phone books, spoken language transcripts, and obtained satisfying segmentation performance.
- 16
- 16
- Symbolic Compression and Processing of Document Images16
A novel encoding scheme is provided that facilitates scalable lossy compression and progressive transmission and supports document image analysis in the compressed domain and a class of document image understanding tasks that operate on the compressed representation.
- 15
- 15
- BiRe-ID: Binary Neural Network for Efficient Person Re-ID15
This work discovers that the significant performance drop of binarized models for Re-ID task is caused by the degraded representation capacity of kernels and features, and proposes the kernel and feature refinement based on generative adversarial learning (KR-GAL and FR- GAL) to enhance the representation capacityof BNNs.
- Rectified Binary Convolutional Networks with Generative Adversarial Learning15
A Rectified Binary Convolutional Network is introduced by combining full precision kernels and feature maps to rectify the binarization process in a generative adversarial network (GAN) framework and further prune the RBCNs using the GAN framework to increase the model efficiency and promote flexibly in practical applications.
- Learning Robot Swarm Tactics over Complex Adversarial Environments15
This paper presents a systematic approach to learn tactical mission-specific policies that compose primitives in a swarm to accomplish the mission efficiently using neural networks with special input and output encoding.
- Cogradient Descent for Bilinear Optimization15
A Cogradient Descent algorithm to address the bilinear problem, based on a theoretical framework to coordinate the gradient of hidden variables via a projection function, which improves the state-of-the-art by a significant margin.
- 15
- 15
- 15
- Mining tools for surveillance video15
A high level query language for submitting queries about spatial and temporal relations of background regions and moving entities, and about human activities and a compiler to map high level queries into a set of novel Petri net filters that utilize computer vision algorithms to answer components of the queries.
- Personalized Large Vision-Language Models14
PLVM proposes Aligner, a pre-trained visual encoder to align referential concepts with the queried images, enabling personalization and the computational cost and parameter count of the Aligner are negligible within the entire framework.
- iffDetector: Inference-Aware Feature Filtering for Object Detection14
This article introduces a generic inference-aware feature filtering (IFF) module that can be easily combined with existing detectors, resulting in the iffDetector, which consistently outperforms state-of-the-art methods with significant margins.
- DOCLIB: a software library for document processing14
A new library called DOCLIB is described, which tries to overcome the drawbacks of earlier approaches and serves the needs of researchers better than previous approaches and will readily be accepted by a larger group of scientists.
- 13
- 13
- 13
- Context-aware and content-based dynamic Voronoi page segmentation13
A dynamic approach to document page segmentation based on inter-component relationships, local patterns and context features is presented and an increase of 15% over the accuracy reported in [2] is achieved.
- 13
- 13
- Desparately seeking Cebuano13
An effort to rapidly develop language resources and component technology to support searching Cebuano news stories using English queries and results from the first 60 hours are presented.
- 13
- Context-Aware Chart Element Detection12
This work proposes a novel method CACHED, which stands for Context-Aware Chart Element Detection, by integrating a local-global context fusion module consisting of visual context enhancement and positional context encoding with the Cascade R-CNN framework.
- 12
- 12
- 12
- 12
- Fast camera motion estimation for hand-held devices and applications12
This paper presents an efficient motion estimation algorithm for camera-enabled handheld devices, such as mobile phones and PDAs, which includes a multi-resolution scheme which searches the match from coarse to fine; and optimization of the search space.
- 12
- Text Extraction, Enhancement and OCR in Digital Video12
Experimental results show that the text extraction scheme can extract both scene text and graphical text robustly and reasonable OCR results are achieved after enhancement.
- 11
- 11
- 11
- Re-targetable OCR with Intelligent Character Segmentation11
This work is a step toward the recognition of scripts of low-density languages which typically do not warrant the development of commercial OCR, yet often have complete TrueType font descriptions.
- 11
- Translation lexicon acquisition from bilingual dictionaries11
This work describes a system for extracting term lexicons from printed copies of bilingual dictionaries and uses the approach to page and definition segmentation and entry parsing to parse a number of dictionaries.
- Residual coding in document image compression11
A set of residual coding models for use with symbolic document image compression that exhibit desirable characteristics for compression and rate-distortion and facilitate compressed-domain processing are described.
- PathDiff: Histopathology Image Synthesis with Unpaired Text and Mask Conditions10
This work proposes PathDiff, a diffusion framework that effectively learns from unpaired mask-text data by integrating both modalities into a unified conditioning space, which allows precise control over structural and contextual features, generating high-quality, semantically accurate images.
- 10
- 10
- Generalization of Hindi OCR Using Adaptive Segmentation and Font Files10
An adaptive Indic OCR system implemented as part of a rapidly retargetable language tool effort and extended, a step toward the recognition of scripts of low-density languages which typically do not warrant the development of commercial OCR, yet often have complete TrueType font descriptions.
- 10
- Shot Boundary Detection using Pixel-to-Neighbor Image Differences in Video10
This paper introduces a framework for shot boundary detection used by the Language and Media Processing Laboratory (LAMP) at University of Maryland, which handles detection of cuts, dissolves and fades.
- 10
- Compressed Video Segmentation10
This paper presents robust techniques for parsing MPEG-compressed video sequences into shots based on their physical structure and further into scenesbased on their semantic structure by identifying changes in content and camera motion.
- One-Shot Learning-Based Handwritten Word Recognition9
This paper explores a one-shot learning approach to recognizing handwritten words using Siamese networks to classify the handwritten images at the word level using a convolutional architecture.
- Mobile Visual Aid Tools for Users with Visual Impairments9
The challenge is how to build a real-time recognition system on mobile devices and the detailed solutions are presented and the pattern recognizer is discussed in detail.
- 9
- 8
- 8
- 8
- 8
- Calibrated Stochastic Gradient Descent for Convolutional Neural Networks8
A theorem is developed to prove that an unbiased estimator for the network variables can be obtained in a probabilistic way based on the Lipschitz hypothesis, and a generic gradient calibration layer is developed which can be easily used to build convolutional neural networks.
- 8
- 8
- Computer vision and image processing techniques for mobile applications8
This dissertation has developed a set of techniques for ego-motion estimation, enhancement, feature extraction, perspective correction, object detection, and document retrieval that serve as a basis for a series of computer vision and image processing techniques together with their applications on the mobile device.
- 8
- Locally adaptive document skew detection8
This paper proposes a new approach to the detection of local orientation and skew in document images based on the observation that there are many documents where a single global estimate of the page skew is not sufficient.
- Instrument grasp: a model and its effects on handwritten strokes8
It is suggested how this information is a valuable heuristic in recovering the direction of motion of the instrument which created the stroke and its application to the use of on-line recognition techniques in processing off-line data is discussed.
- 7
- 7
- 7
- Leveraging Statistical Transliteration for Dictionary-Based English-Bengali CLIR of OCR'd Text7
Experiments with transliteration of out- of-vocabulary English terms into Bengali to improve the effectiveness of English-Bengali Cross-L anguage Information Retrieval and a statistical translation model is used as a basis.
- Correcting errors in digital lexicographic resources using a dictionary7
A paradigm for combining manual and automatic error correction of noisy structured lexicographic data is described and advantages of the paradigm for the task of editing digital bilingual dictionaries are discussed.
- Correcting Errors in Digital Lexicographic Resources Using a Dictionary Manipulation Language7
A paradigm for combining manual and automatic error correction of noisy structured lexicographic data is described and advantages of the paradigm for the task of editing digital bilingual dictionaries are discussed.
- A Novel 2D Marker Design and Application for Object Tracking and Event Detection7
A spatially efficient and robust 2D barcode, M-Code (Marker-Code), which can be attached to the surface of static or moving objects and which not only identifies and tracks an object but also reflects its position and the orientation of the surface where it is attached.
- Imaging as an alternative data channel for camera phones7
A solution to use cameras to down-load data to cell phones as an alternative to existing wireless (CDMA/GPRS, BlueTooth), infrared or cable connections is demonstrated.
- Feature Normalization for Video Indexing and Retrieval7
A frame-type independent representation of the various types of frames present in an MPEG video in which all frames can be considered equivalent is developed.
- Using Physiological Information to Classify Task Difficulty in Human-Swarm Interaction6
The analysis revealed that the temporal lobe in experts (T4 and T3) is predominant for task difficulty classification compared with novices, and showed the importance of the occipital lobe (O1 and O2) coherence feature with the other brain regions.
- Uncertainty-aware Binary Neural Networks6
This work introduces an uncertainty-aware BNN (UaBNN) by leveraging a new mapping function called certainty-sign (c-sign) to reduce these weights' uncertainties, and introduces a c-sign function, the first to train BNNs with a decreasing uncertainty for binarization.
- Proceedings of the 9th IAPR International Workshop on Document Analysis Systems6
This year's workshop is organized as a three-day, single track event, with oral and poster presentations, as well as working group discussions on the second and third afternoons, and special sessions on contributed datasets and a keynote talk on this same topic provide a compelling theme.
- Learning Visual Shape Lexicon for Document Image Content Recognition6
This work presents a new approach for document image content categorization using a lexicon of shape features that is capable to handle high intra-class variability and shows results that exceed other state-of-the-art approaches, allowing it to be used as a content recognizer in image indexing and retrieval systems.
- Simultaneous Appearance Modeling and Segmentation for Matching People Under Occlusion6
An approach to segmenting foreground regions corresponding to a group of people into individual humans by using a nonparametric kernel density estimator in a joint spatial-color space and a recursive probability update scheme is employed for soft segmentation at each iteration.
- 6
- 6
- Content features for logical document labeling6
A model-based approach to combining content features with other geometrical and presentation features for logical labeling in documents without rigid layout structure, like business letters is discussed.
- 6
- 5
- 5
- 5
- LineFormer: Rethinking Line Chart Data Extraction as Instance Segmentation5
This work proposes LineFormer, a robust approach to line data extraction using instance segmentation that achieves state-of-the-art performance on several benchmark synthetic and real chart datasets.
- 5
- SHOE5
A joint optimization framework to learn projections which improve the accuracy of supervised hashing over the current state of the art with respect to standard and sibling evaluation metrics and a sequential greedy optimization that learns relationship aware projections.
- 5
- 5
- Linguistic Resources for Handwriting Recognition and Translation Evaluation5
This work describes efforts to create corpora to support development and evaluation of handwriting recognition and translation technology and develops a stable pipeline and infrastructures for collecting and annotating handwriting linguistic resources to support the evaluation of MADCAT and OpenHaRT.
- TDT-2004: ADAPTIVE TOPIC TRACKING AT MARYLAND5
Results indicate that supervised adaptation and score normalization should be more closely coupled, and that current techniques for detection error tradeoff analysis may be of limited utility when supervised adaptation is performed.
- Video Indexing and Retrieval at UMD.5
An interactive interface that treats each video clip as visual object in a multi-dimensional space, and each ”feature” of that clip is mapped to one dimension and the ability for the users to formulate dynamic queries previously developed for database applications at Maryland is developed.
- Graphical tools and techniques for querying document image databases5
This paper describes document models and relations for the retrieval of document images and the underlying methodology was developed for the Intelligent Document Image Retrieval System (IDIR), which aims to extend document image database query capabilities.
- TRACE: Evidence Grounding-Guided Multi-Video Event Understanding and Claim Generation4
TRACE is introduced, an evidence grounding-guided framework that follows a ground-before-reasoning strategy for multi-video event reasoning and attains state-of-the-art results on the official MAGMaR 2026 leaderboard.
- 4
- 4
- Scene text recognition: an Indic perspective4
A spatial attention-based model (LaSA-Net) that combines visual features and language knowledge for word recognition from scene image word segments is proposed, and a novel language-attunement loss is augmented that enhances the model’s ability to perform zero-shot word recognition.
- Filter Pruning for Efficient CNNs via Knowledge-driven Differential Filter Sampler4
A novel Knowledge-driven Differential Filter Sampler~(KDFS) with Masked Filter Modeling~(MFM) framework for filter pruning, which globally prunes the redundant filters based on the prior knowledge of a pre-trained model in a differential and non-alternative optimization.
- 4
- 4
- Detecting Structural Irregularity in Electronic Dictionaries Using Language Modeling4
A method for dictionary writers to quickly audit structural regularity across entries in a dictionary by using statistical language modeling and calculates the probability of each XML tree in the dictionary against these patterns to look for entries that diverge from the norm.
- 4
- Morphology induction from limited noisy data using approximate string matching4
This work presents algorithms for discovering morphemes from limited, noisy data obtained by scanning a hard copy dictionary, based on the novel application of the longest common substring and string edit distance metrics.
- 4
- Shot boundary detection based on Image correlation features in video4
This paper introduced a framework of shot boundary detection used by the LAMP LAB, University of Maryland, College Park, which deals with cuts and dissolves detection based on image correlation features in the videos.
- Parsing and Tagging of Bilingual Dictionary4
A system for extracting term lexicons from printed bilingual dictionaries by parsing a variety of dictionaries with both Latin and non-Latin alphabets, and demonstrating the results of term lexicon generation for retrieval from a collection of French news stories using English queries.
- 4
- 4
- Model-Based Graphics Recognition4
A novel probabilistic framework for document analysis using an explicit descriptive model of the document class to find the most likely interpretation of a scanned document image, allowing for an end-to-end propagation of the uncertainty.
- Accelerating Social Science Research via Agentic Hypothesization and Experimentation3
An agentic framework that operationalizes end-to-end discovery through a Bayesian optimization inspired two-phase search, in which a Generator proposes candidate hypotheses and an Experimenter evaluates them empirically, which consistently discovers 2-4x more statistically significant hypotheses that are 7-17 percent more predictive than prior approaches.
- CWCD: Category-Wise Contrastive Decoding for Structured Medical Report Generation3
This work proposes Category-Wise Contrastive Decoding (CWCD), a novel and modular framework designed to enhance structured radiology report generation (SRRG), which introduces category-specific parameterization and generates category-wise reports by contrasting normal X-rays with masked X-rays using category-specific visual prompts.
- 3
- 3
- A Survey on Figure Classification Techniques in Scientific Documents3
This survey paper systematically categorizes figures into five classes - tables, photos, diagrams, maps, and plots, and presents a critical review of the existing methodologies and data sets that address the problem of figure classification.
- Anti-Bandit for Neural Architecture Search3
This paper reformulate NAS as a multi-armed bandit problem and presents Anti-Bandit NAS (ABanditNAS) method, which exploits Upper Confidence Bounds (UCB) to abandon arms for search efficiency and Lower Conf confidence bounds (LCB) for fair competition between arms.
- Maryland at FIRE 2011: Retrieval of OCR’d Bengali3
The experiments focused on evaluating a retrieval strategy motivated by recent work on Cross-Language Information Retrieval, but which makes use of OCR error modeling rather than parallel text alignment, and indicate that this is a promising way of using O CR error modeling to improve CLIR.
- 3
- Resource generation from structured documents for low-density languages3
An entry tagging system that recognizes, parses, and tags the entries of a printed dictionary to reproduce the representation and a natural language morphology discovery framework that operates on information from limited, noisy data obtained from the conversion process.
- SOFTCBIR: Object Searching in Videos Combining Keypoint Matching and Graduated Assignment3
This paper proposes a new approach to object searching in video databases, SoftCBIR, which combines a keypoint matching algorithm and a graduated assignment algorithm based on softassign, and proves the e! ectiveness and performance improvements rising from the combination of the two proposed techniques.
- Multimedia applications: beyond similarity searches3
There has been much recent focus on applying machine learning techniques involving semantics modeling, spatio-temporal indexing, multi-modal (audio-, visual-, textual-) integration and relevance feedback searching in Relational database systems.
- 3
- 3
- Compressed-domain document retrieval and analysis3
A data structure is described which can be used to efficiently store the compressed information, provide algorithms for creating and manipulating it, and present results of document processing on images compressed form the University of Washington database.
- CRAFT: Critic-Refined Adaptive Key-Frame Targeting for Multimodal Video Question Answering2
CRAFT (Critic-Refined Adaptive Key-Frame Targeting), a query-conditioned pipeline that combines dynamic keyframe selection, per-video ASR with multilingual fallback, and a hybrid critic loop to iteratively verify and repair claims before consolidation is introduced.
- 2
- 2
- Player Re-Identification Using Body Part Appearences2
A neural network architecture that learns body part appearances for soccer player re-identification that outperforms state-of-the-art models such as OsNet and InceptionNet and yields a robust image-matching feature map.
- 2
- 2
- 2
- JH2R: Joint Homography Estimation for Highlight Removal2
This work addresses the problem of removing unwanted highlight regions in images generated by reflections of light sources on glossy surfaces by using the relationship between the highlight regions resulting in more robust removal of saturated highlights.
- 2
- 2
- Overview of the FIRE 2011 RISOT Task2
The collection used for search was a subset of the FIRE 2008 and 2010 Bengali test collections that contained 92 topics and 62,825 documents.
- Combining Classifiers with Informational Confidence2
The main idea is to adjust confidence values so that their nominal values equal the information actually conveyed, and Shannon’s well-known logarithmic notion of information is used as information measure.
- Adaptive Transformation-Based Learning for Improving Dictionary Tagging.2
An adaptive technique is presented that enables users to produce a high quality dictionary parsed into its lexicographic components (headwords, pronunciations, parts of speech, translations, etc.) using an extremely small amount of user provided training data.
- 2
- 2
- Use of OCR for Rapid Construction of Bilingual Lexicons2
This paper describes an approach to analyzing the lexical structure of OCRed bilingual dictionaries to construct resources suited for machine translation of low-density languages, where online resources are limited.
- Measuring Structural Similarity of Document Pages for Searching Document Image Databases.2
A general system for document image retrieval that is able to make use of document structure is described and a measure of structural similarity between document images based on content area overlap is defined and compared with human relevance judgments.
- 2
- 2
- Data- and content-based adaptation for video over low-bandwidth networks2
This paper focuses on novel techniques to adapt video for its delivery over IP and provides insight into the existing problems and an overview of how video analysis work is being applied in these areas.
- Text quality estimation in video2
This paper addresses the problem of estimating text quality by selecting features based on text resolution, background noise, contrast, illumination and texture, and feeding them into a trained RBF network to estimate the text quality.
- Techniques for the automated testing of document analysis algorithms2
A new approach to automate and manage the testing process for developing document analysis and understanding algorithms and a set of techniques to design test events for test scenarios is proposed aimed at promoting a black-board style research approach.
- 1
- On Measuring Intrinsic Causal Attributions in Deep Neural Networks1
This work treats NNs as structural causal models (SCMs) and extends their focus to include intrinsic causal contributions (ICC) and proposes an identifiable generative post-hoc framework for quantifying ICC.
- 1
- 1
- 1
- A Survey on Change Detection Techniques in Document Images1
A survey on core techniques and rules to detect changes in different versions of a document image, and discussions on change detection focus on two categories -- content-based and layout-based.
- 1
- Confidence Dimension for Deep Learning based on Hoeffding Inequality and Relative Evaluation1
This paper proposes to use multiple factors to measure and rank the relative generalization of DNNs based on a new concept of confidence dimension (CD), and provides a feasible framework in this CD to theoretically calculate the upper bound of generalization based on the conventional Vapnik-Chervonenk dimension (VC-dimension) and Hoeffding's inequality.
- 1
- 1
- 1
- 1
- 1
- COMPUTER VISION AND IMAGE PROCESSING LARGE TECHNIQUES FOR MOBILE APPLICATIONS1
A set of techniques for ego-motion estimation, enhancement, feature extraction, perspective correction, object detection, and document retrieval that serve as a basis for a series of computer vision and image processing techniques together with their applications on the mobile device.
- A Platform Independent Image and Video Engine1
A high level overview of design, a general description of the functionality of various components such as MediaAlerts, MediaCapture, MediaStorage and MediaProcessing are provided and several applications in progress, which utilize media engine components are described.
- 1
- 1
- State of The Art: Informedia Digital Video Library1
In its efforts to create an automatically processed and large-scale video collection, Informedia is not as ambitious, in terms of the content analysis, as the technologies “on the horizon” described in the previous chapter.
- Document Image Coding for Processing and Retrieval1
A coding methodology is described that not only exploits component-level redundancy to reduce code length but also supports efficient data access to facilitate enhanced compression and retrieval of document images.
- Document Understanding - 19961
This report contains nearly 340 references which are directly related to the field of document image understanding and appeared in major journals and conferences during 1996.
- 1
- –
- Attribute Retrieving for Open-Vocabulary Endoscopic Compositional Referring Segmentation–
The Attribute Retrieval-based Endoscopic-RIS (AR-ERIS) framework for open-vocabulary endoscopic compositional referring segmentation leverages attribute retrieval for open-vocabulary endoscopic compositional referring segmentation and is pretrained on the curated ReferEndoscopy dataset.
- –
- –
- –
- –
- Volumetric Directional Diffusion: Anchoring Uncertainty Quantification in Anatomical Consensus for Ambiguous Medical Image Segmentation–
Volumetric Directional Diffusion is proposed, a prior-anchored diffusion framework that shifts stochastic generation from full-mask synthesis to residual boundary exploration and suggests that prior-anchored residual diffusion can model clinically plausible expert disagreement without sacrificing anatomical fidelity.
- –
- ZIPP:Zero-shot Image Personalization from Personas–
This work introduces zero-shot image personalization from personas (ZIPP), which conditions image generation on natural-language personas (concise descriptors of a user's identity and aesthetic sensibilities) without any user-specific data or weight updates.
- Score-Control for Hallucination Reduction in Diffusion Models–
A Variance-Guided Score Modulation (VSM) strategy that controls the score Jacobian, in turn reducing score smoothness and better approximating the ground truth score that decreases hallucinations is introduced, providing a principled step toward more reliable diffusion-based image generation.
- FairLLaVA: Fairness-Aware Parameter-Efficient Fine-Tuning for Large Vision-Language Assistants–
FairLLaVA is introduced, a parameter-efficient fine-tuning method that mitigates group disparities in visual instruction tuning without compromising overall performance and provides an architecture-agnostic approach to fair visual instruction following.
- –
- –
- –
- –
- –
- –
- –
- –
- The Analysis and Extraction of Structure from Organizational Charts–
An automated and end-to-end approach that uses computer vision, deep learning, and natural language processing techniques to solve the problem of manually extracting information from hierarchical chart analysis is presented.
- A Knowledge Enforcement Network-Based Approach for Classifying a Photographer’s Images–
Experimental results of the proposed model show that the method outperforms the existing techniques by 5%–10% on average, and the dataset created for the experimental purpose will be made available upon publication.
- –
- –
- –
- Planar content selection in images and videos using frontalness–
This paper addresses the problem of selecting instances of a planar object in a video or from a set of images based on an evaluation of its “frontalness” by computing how close the object’s surface normal aligns with the optical axis of a camera by exploiting the objective space error.
- –
- –
- –
- –
- Efficient machine methods for document image analysis–
This thesis presents a data selection method for training Support Vector Machines (SVM) on large-scale data sets, and presents a graph-based method for segmentation of handwritten document images into text-lines which is more efficient and adaptive than previous approaches.
- Adaptive algorithms for automated processing of document images–
This work proposes a generic model to segment and to recognize characters for any complex syllabic or non-syllabic script, using font-models, based on the fact that font files contain all the information necessary to render text and thus a model for how to decompose them.
- Content recognition and context modeling for document analysis and retrieval–
A novel multi-scale framework to detect and segment signatures jointly from document images, based on the structural saliency under a signature production model is proposed, and a model-based approach for extracting relevant named entities from unstructured documents is presented.
- Adaptive analysis and processing of structured multilingual documents–
Experimental results working on different scripts, such as Chinese, Korean, Arabic, Devanagari, and Khmer, demonstrate that the proposed framework can save human efforts significantly by making each phase adaptive.
- –
- Review: Content Structuring In Image & Video Repositories–
This chapter provides insight on content management for image and video resources by discussing the content structuring and concept matching facilities commonly used to provide increasing levels of deep semantic concept association to image andVideo objects.
- Conclusions and Technology Directions–
This study has confirmed the mindset of the InfoSleuth approach in CBAM type applications and identified two opportunities to detect content-based concepts as patterns of content abstractions across multiple information sources and over temporal histories.
- Review: Content-Based Access to Text Resources–
This chapter discusses the content structuring and concept matching facilities commonly used to provide increasing levels of deep semantic concept association to textual documents.
- Network-Diffused Media Scaling for Multimedia Content Services–
The approach aims to build an intelligent content service infrastructure in a way that the servers, nodes, and terminals are aware of their capabilities along with the capabilities of their surroundings.
- Technical Issues and Criteria–
The major issues and components to be considered when surveying technologies for distributed and content-based information networks and the information access paradigm well-suited to these environments are set forth.
- Review: Intermedia Trends in The Web–
The current state of the art requires that the content markup facilities that enable semantic rewriting, multi-source collecting, and rank merging across a diverse set of multimedia sources be focused on.
- Review: Media Enhanced input/output Metaphors–
To access multimedia information by content, various query mechanisms need to be combined, and the user interface should be visual as much as possible and also enable visual relevance feedback, user-guided navigation and user-controlled discovery of new associations.
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-15. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.