Feature extraction is an essential step in pattern classification, which is normally divided into two tasks: transforming the input vector into a feature vector and/or reducing its dimensionality. A well-defined feature extraction algorithm makes the subsequent classification process more effective and efficient. One of the most important feature extraction algorithms is linear discriminant analysis (LDA). However, there is a critical drawback for LDA. For a classification task with c classes, since the rank of the between class matrix cannot be larger than c - 1, the dimension of the projected subspace is at most c - 1 for LDA. From this viewpoint, min-max discriminant analysis based on gradient method (MMDA-GM) is derived in this paper. With the proposed MMDA-GM, a set of features can be extracted simultaneously. It is shown that the proposed method achieves good performance for data sets from UCI Machine Learning Repository.

Spatial Pyramid Matching (SPM) has become a standard in bag-of-words (BoW) image representation, regardless of the features used in the process. With most research focusing on other part of the BoW pipeline, the arrangement of spatial windows in SPM remains untouched except by few works. This paper takes into consideration the idea that not all spatial windows in SPM are needed, and proposes two systematic ways of learning the optimal spatial window arrangement, called greedy-OA and linear-OA. In both algortihms, the spatial windows of a SPM pyramid are examined in detail, and optimized to achieve the best performance with the lowest number of spatial windows possible. Using overlapping spatial windows from out previous work to enhance the discriminative power of each spatial windows, our experiments shows that the complete SPM-pyramid is indeed sub-optimal, as we are able to consistently achieve better performance (up to 4. 38%) with a reduced memory cost of 40%. Both algorithm are tested using 15 Scene, Caltech 101, and Caltech 256 datasets.

Image representation holds a very important role in object recognition, as the choice of image representation directly affects the training process and recognition rate. Among many representation technique, Spatial Pyramid Matching (SPM) has been a major breakthrough in the field of object and scene recognition. Expanding from the normal bag-of-words approach, an image is viewed as a pyramid of L + 1 level, where the lthlayer of a pyramid is divided into 2l× 2ldisjoint sub-windows, and a histogram is extracted from each. In SPM, each image is represented by the concatenation of such histograms. This paper proposes an extension to the normal SPM framework by introducing overlapping sub-windows to replace the disjoint sub-windows of SPM. Two frameworks are proposed in an effort to increase discriminability: rectangular overlapping windows (OW) and circular overlapping windows (CW). We found that the introduction of overlapping sub-windows led to better performances in Caltech 101, Caltech 256 and 15-Scene databases by up to 3.68% under ScSPM and LLC algorithm. Furthermore, it was found that using only the 2ndoverlapping layer directly enables us to reduce the memory consumption by 24% while still achieving better recognition rate than traditional SPM.

In this paper, the gradient vector flow fields are introduced in image restoration. Within the context of flow fields, the shock filter, mean curvature flow, and Perona-Malik equation are reformulated. Many advantages over the original models can be obtained; these include numerical stability, large capture range, and high-order derivative estimation. In addition, a fairing process is introduced in the anisotropic diffusion, which contains a fourth-order derivative and is reformulated as the intrinsic Laplacian of curvature under the level set framework. By applying this fairing process, the shape boundaries will become more apparent. In order to overcome numerical errors, the intrinsic Laplacian of curvature is computed from the gradient vector flow fields instead of the observed images.

Robust real-time tracking of non-rigid objects in a dynamic environment is a challenging task. Among various cues in tracking, color can provide an efficient visual cue for this type of tracking problem because of its invariance in the presence of changing complex shapes and appearances. To track the color object, a particle filter uses several hypotheses simultaneously and weights them by their similarity with the object color model. The use of particle filter allows better handling of color clutter and maintains track even through momentary occlusions. However, one of the major difficulties of color-based tracking is that the color changes in dynamic lighting conditions. Different from other color model adaptation methods, we propose an adaptive color-based particle filter tracking algorithm by transductive inference. The particles are given different weights as weighted unlabeled data by a weak classifier. Combining confidently labeled data and weighted unlabeled data, the proposed transductive particle filter algorithm offers an effective way to transduce object color model through the given observation in non-stationary color distributions. Experiments show that the transductive particle filter algorithm can effectively update the object color model and maintain object tracking even under temporal occlusion and color clutter.

In this paper, we address the problem of tracking multiple people in a cluttered environment based on thorough usage of color information. We introduce color invariants to the proposed combined color spatial model, which combines the low-level and middle-level color spatial information together. This formalization gives us a way to combine low-level and middle-level color spatial distribution to build a person model. In multiple targets cluttered environment, color invariant and color spatial information play an important role in robust surveillance system. We propose a modified mean shift algorithm to track multiple people based on color invariants and color spatial model. The interactions of multiple people have also been studied in order to guide the tracking under occlusion.

This paper addresses the issue of color model learning and adaptation when color is used as a feature for object tracking in a dynamic scene. Under different environmental conditions, e.g. illumination changes or non-stationary scenes, a static color model is inadequate and color model adaptation is required. The color model adaptation for object tracking can be classified as an unsupervised (or semi-supervised) learning problem. The algorithm should be able to select the reliable training samples to update the color model automatically. A Bilateral Learning (BL) approach is proposed in this paper. The spatial and color information of the target are combined in the color model adaptation and color based object tracking procedure. The color model and spatial model are adapted in the color space and image space alternatively, which results in the color model adaptation and the localization of the target along the image sequence. Experimental results show the effectiveness and efficacy of the proposed method for the color model adaptation and object tracking under illumination changes and environmental noises.

This paper presents a view invariant face recognition system based on both 3D range data as well as 2D gray-level facial images. An irregular 2D mesh labeled by twelve landmarks and 3D regions labeled by four landmarks are defined for each model for feature extraction. Nodes of the 2D mesh are described by Gabor filter responses and 3D points are represented by point signature. To classify test faces under varying poses from one stored view, a robust structural Hausdorff distance is proposed for non-point-to-point matching under the structural constraints. The best matched model is determined based on the linear integration of matching results in 2D and 3D domains. Experimental results based on our database involving 80 persons with different facial expression and different viewpoint have demonstrated the efficiency of our algorithm.

Current research efforts about passive sensing of 3D human hand pose have been hampered by the computational complexity caused by inverse kinematics and 3D reconstruction. We focuses on 3D hand posture estimation based on a single 2D image. We analyze some of its constraints to reduce the 27 DOFs to 12 DOFs without any significant degradation of performance. A novel algorithm to estimate the 3D hand posture from eight 2D projected feature points is proposed. Experimental results confirm that our algorithm gives good estimates of the 3D hand pose.

We present a novel face recognition algorithm based on the point signature-a representation for free-form surfaces. We treat the face recognition problem as a non-rigid object recognition problem. The rigid parts of the face of one person are extracted after registering the range data sets of faces having different facial expressions. These rigid parts are used to create a model library for efficient indexing. For a test face, models are indexed from the library and the most appropriate models are ranked according to their similarity with the test face. Verification of each model face can be quickly and efficiently identified. Experimental results with range data involving six human subjects, each with four different facial expressions, have demonstrated the validity and effectiveness of our algorithm.

With the growing importance of the robotic manipulators in the area of industrial automation, there is an increase in demand for their position accuracy. To control such robotic systems effectively, more complex forms of control than the simple constant gain linear feedback control is required. Two adaptive control schemes, namely the adaptive PID controller and the self-tuning controller, are implemented in the system. To provide for fast control actions in the robotic system, a real-time approach is adopted. Such operation requires large computing power in a relatively short period of time. A digital signal processor (DSP) whose architecture is optimized to process sampled data at high rate is well suited to this purpose. Its architecture is designed to exploit the repetitive nature of signal processing by pipelining the data flow for extra speed. The TMS320C25 DSP chip is chosen for this purpose. From extensive experimental testing, it is shown that the position accuracy of the robotic manipulator using the given adaptive control schemes is within the 0.5 mm range.>