In a recent work, the effectiveness of neighborhood supported model level fuzzy aggregation was shown under dynamic background conditions. The multi-feature fuzzy aggregation used in that approach uses real fuzzy similarity values, and is robust for low and medium-scale dynamic background conditions such as swaying vegetation, sprinkling water, etc. The technique, however, exhibited some limitations under heavily dynamic background conditions, as features have high uncertainty under such noisy conditions and these uncertainties were not captured by real fuzzy similarity values. Our proposed algorithm is particularly focused toward improving the detection under heavy dynamic background conditions by modeling uncertainties in the data by interval-valued fuzzy set. In this paper, real-valued fuzzy aggregation has been extended to interval-valued fuzzy aggregation by considering uncertainties over real similarity values. We build up a procedure to calculate the uncertainty that varies for each feature, at each pixel, and at each time instant. We adaptively determine membership values at each pixel by the Gaussian of uncertainty value instead of fixed membership values used in recent fuzzy approaches, thereby, giving importance to a feature based on its uncertainty. Interval-valued Choquet integral is evaluated using interval similarity values and the membership values in order to calculate interval-valued fuzzy similarity between model and current. Adequate qualitative and quantitative studies are carried out to illustrate the effectiveness of the proposed method in mitigating heavily dynamic background situations as compared to state-of-the-art.

Object tracking from video sequences, although well researched, still faces difficulties under certain challenging situations. We propose a new particle filter-based tracking, where the fuzzy observation model exploits spatial correlation in a rough set-theoretic framework. We further improved certain critical steps of the particle filter such as mean state estimation and resampling. In our approach, mean state of the object at each time step is estimated with an adaptive number of high-weighted samples. These high-weighted samples are propagated along with the fresh samples generated around the current mean state to maintain sample diversity and avoid sample impoverishment. We later extended our approach to multi-cues. Given the wide applicability of statistical texture in various vision applications due to its robustness under various challenging situations, we propose to use fuzzy statistical texture cues along with fuzzy roughness color cues, and the weight for each cue is adaptively determined using the discriminating ability of the cue. Results show that our proposed particle filtering approach has less error variance across simulations and can perform tracking under challenging situations better than the contemporary ones.

In this paper, we examine the suitability of correlogram for background subtraction, as a step towards moving object detection. Correlogram captures inter-pixel relationships in a region and is seen to be effective for modeling the dynamic backgrounds. A multi-channel correlogram is proposed using inter-channel and intra-channel correlograms to exploit full color information and the inter-pixel relations on the same color planes and across the planes. We thereafter derive a novel feature, termed multi-channel kernel fuzzy correlogram, composed by applying a fuzzy membership transformation over multi-channel correlogram. Multi-channel kernel fuzzy correlogram maps multi-channel correlogram into a reduced dimensionality space and is less sensitivity to noise. The approach handles multimodal distributions without using multiple models per pixel unlike traditional approaches. The approach does not require ideal background frames for background model initialization and can be initialized with moving objects also. Effectiveness of the proposed method is illustrated on different video sequences.

We propose a new algorithm for moving object detection in the presence of challenging dynamic background conditions. We use a set of fuzzy aggregated multifeature similarity measures applied on multiple models corresponding to multimodal backgrounds. The algorithm is enriched with a neighborhood-supported model initialization strategy for faster convergence. A model level fuzzy aggregation measure driven background model maintenance ensures more robustness. Similarity functions are evaluated between the corresponding elements of the current feature vector and the model feature vectors. Concepts from Sugeno and Choquet integrals are incorporated in our algorithm to compute fuzzy similarities from the ordered similarity function values for each model. Model updating and the foreground/background classification decision is based on the set of fuzzy integrals. Our proposed algorithm is shown to outperform other multi-model background subtraction algorithms. The proposed approach completely avoids explicit offline training to initialize background model and can be initialized with moving objects also. The feature space uses a combination of intensity and statistical texture features for better object localization and robustness. Our qualitative and quantitative studies illustrate the mitigation of varieties of challenging situations by our approach.

Distributed video coding (DVC) suits the light weight encoder and complex decoders, shifting the motion estimation to the decoder, in contrast to the conventional encoding scheme. Mode decision is one of the factor affecting the Rate Distortion (RD) performance due to the varying statistics of the sequence in the DVC. In this paper, we propose to introduce coding modes and the Side Information (SI) refinement modes at the decoder in the pixel domain multi resolution refinement(MRR) Wyner-Ziv scheme. During the decoding process, the mode decision of the current decoding layer of the block is based on the information from the previous decoded layers of the block and the updated SI. We further propose the adaptive weighted reconstruction scheme using the neighbor pixels. RD performance of the proposed scheme is tested with several sequences. The results shows significant performance improvement.

It is seen that the performance of a video codec drastically degrades for noisy sources. In this paper, we propose a novel and easy-to-implement de-noising filter, integrated within the motion estimation and motion compensation blocks of an H.264 video encoder, to reduce impulse noise from video. Performance of our proposed filter is compared with a spatial domain two-pass adaptive median filter that uses spatial information of noise. Results are obtained for several video sequences, contaminated with different levels of additive random noise, thus establishing the effectiveness of the proposed technique.

In this paper, we present a novel approach towards customized and automated generation of sports highlights from its extracted events and semantic concepts. A recorded sports video is first divided into slots, based on the game progress and for each slot, an importance-based concept and event-selection is proposed to include those in the highlights. Using our approach, we have successfully extracted highlights from recorded video of cricket match

We present a genetic algorithm based integrated approach to model-assisted matching and pose estimation for automated visual inspection applications. Unlike the past works reported in literature, this approach does not consider the matching between the model and the image of the object to be essential step prior to pose estimation. A set of matched vertices sequence and poses are hypothesized using a newly proposed composite chromosome structure and these are genetically evolved until a reasonably accurate pose is determined. Our algorithm demonstrates its robustness against noise as well as missing and spurious object vertices.

This work presents a new perceptually motivated bit allocation strategy for region of interest (ROI) coding in video sequences. It is not possible to assure any minimum target foreground quality in the approaches reported so far in the literature. In our proposed strategy reported herein, the bit-allocation is based on a target quality for the foreground and allowing perceptually tolerable graceful quality degradation for the background within the specified bit-rate constraint. In cases where it is difficult to satisfy the target foreground quality within the bit rate constraints, the closest possible quality is assured. Our algorithm is implemented on the H.264/AVC standards. Results on some standard video sequences are presented and these are very encouraging.

In this paper, we present an accurate and robust pose estimator of rigid, polyhedral objects, based on Artificial Neural Networks (ANN), as suitable for Automated Visual Inspection (AVI) applications. The estimator is novel in the sense that it is trained with different poses of the objects having dimensional deviations within its tolerance range and is therefore robust with respect to within tolerance dimensional errors. The estimation accuracy is scalable and our computer simulation experiments in the existing configurations of ANNs have shown an accuracy better than 4% of the placement error. The ANN based pose estimator offers several advantages over the classical implementations.

A novel approach to selectively enhance important facial features in very low bit-rate head and shoulders video sequences is described. The selected facial features (eyes, nose and mouth) are tracked from frame to frame and their positions are used to rank image blocks according to their perceptual importance. This ranked list of blocks is used to generate an independent bit stream which can be used to enhance regions of interest in the reconstructed image, depending on the available transmission bit rate. The technique is demonstrated as an add-on to an available H.263 encoder and decoder. Results are presented which show the effectiveness of the method for common image sequences.

A novel approach to selectively enhancing important facial features in head and shoulders video sequences is described. The selected facial features (eyes, nose and mouth) are tracked from frame to frame and their positions are used to rank image blocks according to their perceptual importance. This ranked list of blocks is used to generate an independent bit stream which can be used to enhance regions of interest in the reconstructed image, depending on the available transmission bit rate. The technique is demonstrated as an add-on to an available JPEG-2000 encoder and decoder. Results are presented which show the effectiveness of the method for common image sequences. The applicability of this technique to inter-frame encoded motion compensated video is noted.