Showing papers by "Santanu Chaudhury published in 2002"

Devnagari numeral recognition by combining decision of multiple connectionist classifiers

Abstract: This paper is concerned with recognition of handwritten Devnagari numerals. The basic objective of the present work is to provide an efficient and reliable technique for recognition of handwritten numerals. Three different types of features have been used for classification of numerals. A multi-classifier connectionist architecture has been proposed for increasing reliability of the recognition results. Experimental results show that the technique is effective and reliable.

Developing a Gesture-based Interface

Abstract: A gesture-based interface involves tracking a moving hand across frames, and extracting the semantic interpretation corresponding to the gesture. This is a difficult task, since there is a change in both the position as well as the appearance of the hand. Further, such a system should be robust to the speed at which the gesture is performed. This paper presents a novel attempt at developing a hand gesture-based interface. We propose an on-line predictive EigenTracker for the moving hand. Our tracker can learn the eigenspace on the fly. We propose a new state-based representation scheme for hand gestures, based on the eigenspace reconstruction error. This makes the system independent of the speed of performing the gesture. We use learning for adapting the gesture recognition system to individual requirements. We show results of successful operation of our system even in cases of background clutter and other moving objects.

Generic Video Classification: An Evolutionary Learning Based Fuzzy Theoretic Approach.

Abstract: In this paper we propose an evolutionary learning based fuzzy theoretic approach for classifying video sequences into generic categories. This categorization is based on video structure based syntactic features. The features like shot durations, editing style, camera work and shot activity conveys large amount of information about the type of video. The information derived from these features is integrated over a larger time-scale than a shot length time to form fuzzy rules for the extraction of video structure based features. Evolutionary learning paradigm is used to evolve the fuzzy rule based system for generic video characterization. Such a rule-based system yields high representational accuracy of the classes as shown by the experiments conducted on various type of video sequences ranging up to 1 to 3 minutes. Experimental analysis illustrates the effectiveness of our system in offering a novel approach for categorizing the video sequences.

Enhanced Video Representation Using Objects.

Abstract: Object-based video representations such as MPEG-4 have opened up new possibilities for video content access and manipulation. In this paper, we present a new objectbased video representation paradigm, using appearance spaces. Our scheme enables fully automated extraction of semantic video objects for a class of sequences, and the development of a compressed, highly flexible representation of the video sequence based on extracted content. Our video representation supports content-based retrieval, as well as numerous enhanced features such as hyperlinking of videos, motion-icons for video browsing, automatic annotation, content-authoring facilities and se-

CONDENSATION-Based Predictive EigenTracking.

Abstract: An appearance-based EigenTracker can track objects which simultaneously undergo image motions as well as changes in view This paper enhances the framework in two ways First, we incorporate a novel CONDENSATION-based predictive framework to speed up the EigenTracker Next, our scheme is on-line: we use efficient eigenspace updates to track unknown objects We use Importance Sampling for enhancing the capability of an EigenTracker Our on-line EigenTracker is flexible it is possible to use it in conjunction with other trackers in a symbiotic manner We show results of efficient and successful tracking for two important applications – hand gesture analysis, and face and person tracking

An appearance based approach for video object extraction and representation

Abstract: We describe a novel appearance based scheme for extraction and representation of video objects. The tracking algorithm used for video object extraction is based upon a new eigen-space update scheme. We propose a scheme for organisation of video objects in an appearance based hierarchy. The appearance based hierarchy is constructed using a new SVD based eigen-space merging algorithm. The hierarchy enables approximate query resolution. Experiments performed on a large number of video sequences have yielded promising results.

Semantic Categorization of Video: An Evolutionary Learning based Approach

Abstract: In this paper, we have proposed a fuzzy rule based system for classification of video into semantic categories. The classification scheme uses an evolutionary learning methodology to evolve a fuzzy system for use in the classification process. This evolved fuzzy classifier has the inherent capability to tackle variations and ambiguities invariably present in the video data. A novel fuzzy theoretic sheme has been suggested for extraction of key frames from a given video after shot segmentation. Frame based temporal features and spatial features obtained from key frames have been used in the classification system. We have developed an experimental system for categorization of sports video. The experimental system has yielded reasonably correct recognition results for a large number of samples.