Sethuraman Panchanathan

Review of Image and Video Indexing Techniques

Visual database systems require efficient indexing to facilitate fast access to the images and video sequences in the database. Recently, several content-based indexing methods for image and video based on spatial relationships, color, texture, shape, sketch, object motion, and camera parameters have been reported in the literature. The goal of this paper is to provide a critical survey of existing literature on content-based indexing techniques and to point out the relative advantages and disadvantages of each approach.

Error Resiliency Schemes in H.264/AVC Standard

Abstract Real-time transmission of video data in network environments, such as wireless and Internet, is a challenging task, as it requires high compression efficiency and network-friendly design. H.264/AVC is the newest international video coding standard, jointly developed by groups from ISO/IEC and ITU-T, which aims at achieving improved compression performance and a network-friendly video representation for different types of applications, such as conversational, storage, and streaming. In this paper, we discuss various error resiliency schemes employed by H.264/AVC. The related topics such as non-normative error concealment and network environment are also described. Some experimental results are discussed to show the performance of error resiliency schemes.

A critical evaluation of image and video indexing techniques in the compressed domain

Abstract Image and video indexing techniques are crucial in multimedia applications. A number of indexing techniques that operate in the pixel domain have been reported in the literature. The advent of compression standards has led to the proliferation of indexing techniques in the compressed domain. In this paper, we present a critical review of the compressed domain indexing techniques proposed in the literature. These include transform domain techniques using Fourier transform, cosine transform, Karhunen–Loeve transform, Subbands and wavelets, and spatial domain techniques using vector quantization and fractals. In addition, temporal indexing techniques using motion vectors are also discussed.

Biased Manifold Embedding: A Framework for Person-Independent Head Pose Estimation

The estimation of head pose angle from face images is an integral component of face recognition systems, human computer interfaces and other human-centered computing applications. To determine the head pose, face images with varying pose angles can be considered to be lying on a smooth low-dimensional manifold in high-dimensional feature space. While manifold learning techniques capture the geometrical relationship between data points in the high-dimensional image feature space, the pose label information of the training data samples are neglected in the computation of these embeddings. In this paper, we propose a novel supervised approach to manifold-based non-linear dimensionality reduction for head pose estimation. The Biased Manifold Embedding (BME) framework is pivoted on the ideology of using the pose angle information of the face images to compute a biased neighborhood of each point in the feature space, before determining the low-dimensional embedding. The proposed BME approach is formulated as an extensible framework, and validated with the Isomap, Locally Linear Embedding (LLE) and Laplacian Eigen-maps techniques. A Generalized Regression Neural Network (GRNN) is used to learn the non-linear mapping, and linear multi-variate regression is finally applied on the low-dimensional space to obtain the pose angle. We tested this approach on face images of 24 individuals with pose angles varying from -90deg to +90deg with a granularity of 2. The results showed substantial reduction in the error of pose angle estimation, and robustness to variations in feature spaces, dimensionality of embedding and other parameters.

Automated gesture segmentation from dance sequences

Complex human motion (e.g. dance) sequences are typically analyzed by segmenting them into shorter motion sequences, called gestures. However, this segmentation process is subjective, and varies considerably from one choreographer to another. Dance sequences also exhibit a large vocabulary of gestures. In this paper, we propose an algorithm called hierarchical activity segmentation. This algorithm employs a dynamic hierarchical layered structure to represent human anatomy, and uses low-level motion parameters to characterize motion in the various layers of this hierarchy, which correspond to different segments of the human body. This characterization is used with a naive Bayesian classifier to derive choreographer profiles from empirical data that are used to predict how particular choreographers segment gestures in other motion sequences. When the predictions were tested with a library of 45 3D motion capture sequences (with 185 distinct gestures) created by 5 different choreographers, they were found to be 93.3% accurate.

Fast Wavelet Histogram Techniques for Image Indexing

Content-based image indexing is emerging as an important research area with application to digital libraries and multimedia databases. A majority of indexing techniques are based on pixel domain features such as histogram, color, texture, and shape. However, with recent advances in image compression, compressed domain indexing techniques are gaining popularity due to their low complexity. Recently, a wavelet histogram technique which exploits the directional properties of wavelet transform has been proposed in the literature. Although, this technique provides a good retrieval performance for texture images, its complexity is very high. In this paper, we propose three techniques to reduce the complexity of the wavelet histogram method. These techniques together provide a superior performance at a substantially reduced complexity and, hence, can be considered as a potential candidate for developing a joint wavelet-based image storage and retrieval system.

Experiments on block-matching techniques for video coding

Video compression is becoming increasingly important with the advent of compression standards and broadband networks. Recently, several block-based motion-estimation algorithms to exploit the temporal redundancies in a video sequence have been reported in the literature. Some of these algorithms tend to be either computationally expensive or to converge to a local optimum. In this paper, we present results for various block-matching techniques and propose a low-complexity block-matching motion-estimation algorithm that is useful for hybrid video coding schemes, including MPEG video. This algorithm consists of a layered structure search, and, unlike other fast block-matching methods, it does not converge to a local optimum. The proposed method employs a novel matching criterion, namely, the modified pixeldifference classification (MPDC), that offers simplicity with other potential advantages.

Gesture segmentation in complex motion sequences

Complex human motion sequences (such as dances) are typically analyzed by segmenting them into shorter motion sequences, called gestures. However, this segmentation process is subjective, and varies considerably from one human observer to another. In this paper, we propose an algorithm called hierarchical activity segmentation. This algorithm employs a dynamic hierarchical layered structure to represent the human anatomy, and uses low-level motion parameters to characterize motion in the various layers of this hierarchy, which correspond to different segments of the human body. This characterization is used with a naive Bayesian classifier to derive creator profiles from empirical data. Then those profiles are used to predict how creators will segment gestures in other motion sequences. When the predictions were tested with a library of 3D motion capture sequences, which were segmented by 2 choreographers they were found to be reasonably accurate.

A virtual reality simulator for orthopedic basic skills: A design and validation study

Orthopedic drilling as a skill demands high levels of dexterity and expertise from the surgeon. It is a basic skill that is required in many orthopedic procedures. Inefficient drilling can be a source of avoidable medical errors that may lead to adverse events. It is hence important to train and evaluate residents in safe environments for this skill. This paper presents a virtual orthopedic drilling simulator that was designed to provide visiohaptic interaction with virtual bones. The simulation provides a realistic basic training environment for orthopedic surgeons. It contains modules to track and analyze movements of surgeons, in order to determine their surgical proficiency. The simulator was tested with senior surgeons, residents and medical students for validation purposes. Through the multi-tiered testing strategy it was shown that the simulator was able to produce a learning effect that transfers to real-world drilling. Further, objective measures of surgical performance were found to be able to differentiate between experts and novices.

Analysis of low resolution accelerometer data for continuous human activity recognition

The advent of wearable sensors like accelerometers has opened a plethora of opportunities to recognize human activities from other low resolution sensory streams. In this paper we formulate recognizing activities from accelerometer data as a classification problem. In addition to the statistical and spectral features extracted from the acceleration data, we propose to extract features that characterize the variations in the first order derivative of the acceleration signal. We evaluate the performance of different state of the art discriminative classifiers like, boosted decision stumps (AdaBoost), support vector machines (SVM) and regularized logistic regression (RLogReg) under three different evaluation scenarios (namely subject independent, subject adaptive and subject dependent). We propose a novel computationally inexpensive methodology for incorporating smoothing classification temporally, that can be coupled with any classifier with minimal training for classifying continuous sequences. While a 3% increase in the classification accuracy was observed on adding the new features, the proposed technique for continuous recognition showed a 2.5 - 3% improvement in the performance.