Catálogo de publicaciones - libros

This paper presents a system that tracks human heads in real-time under unconstrained environments where target occlusion, varying illumination, and cluttered backgrounds exist. Tracking is formulated as an active visual servo problem based on the integration of a saccade and a smooth pursuit processes. The head is modelled as an ellipse computed from the color clusters of candidate targets using a robust least square ellipse fitting algorithm. The Farnsworth Perceptually Uniform Color Model is employed to represent the color information of the visual objects. Kalman filtering is applied to the head ellipse to track the evolution of the position, size, and orientation of the target such that the occlusion of objects with similar color and shape as those of the target are effectively accommodated. Experiments with tracking scenarios demonstrate the effectiveness of the system.

A novel region-based multiple object tracking framework based on Kalman filtering and elastic matching is proposed. The proposed Kalman filtering-elastic matching model is general in two significant ways. First, it is suitable for tracking of both, rigid and elastic objects. Second, it is suitable for tracking using both, fixed cameras and moving cameras since the method does not rely on background subtraction. The elastic matching algorithm exploits both the spectral features and structural features of the tracked objects, making it more robust and general in the context of object tracking. The proposed tracking framework can be viewed as a generalized Kalman filter where the elastic matching algorithm is used to measure the velocity field which is then approximated using B-spline surfaces. The control points of the B-spline surfaces are directly used as the tracking variables in a grid-based Kalman filtering model. The limitations of the Gaussian distribution assumption in the Kalman filter are overcome by the large capture range of the elastic matching algorithm. The B-spline approximation of the velocity field is used to update the spectral features of the tracked objects in the grid-based Kalman filter model. The dynamic nature of these spectral features are subsequently used to reason about occlusion. Experimental results on tracking of multiple objects in real-time video are presented.

Singular (unobservable) movements pose major challenges for consistent 3D human arm tracking using monocular image sequences. In this paper, we present an efficient and robust method for the detection and tracking recovery from one of the singular movements: rotation about humerus with outstretched arm. In our approach using a particle filter for 3D arm tracking, movement constraints (i.e. range of arm joint angles) are not enforced in particle generation. Instead, singularity detection is achieved by looking for particles with joint angles violating these constraints. Once such a singular movement has been detected, inverse kinematics method is used to recover correct arm tracking by transferring invalid particles from unconstrained movement parameter space into valid constrained space. Experimental results have demonstrated the efficacy of our approach in terms of explicit singularity detection, fast recovery of tracking and small number of particles.

In multi-target visual tracking, tracking failure due to miss-association can often arise from the presence of occlusions between targets. To cope with this problem, we propose the predictive estimation method that iterates occlusion prediction and occlusion status update using occlusion activity detection by utilizing joint probabilistic data association filter in order to track each target before, during and after occlusion. First, the tracking system predicts the position of a target, and occlusion activity detection is performed at the predicted position to examine if an occlusion activity is enabled. Second, the tracking system re-computes positions of occluded targets and updates them if an occlusion activity is enabled. Robustness of multi-target tracking using predictive estimation method is demonstrated with representative simulations.

A better understanding of cell behavior is very important in drug and disease research. Cell size, shape, and motility may play a key role in stem-cell specialization or cancer development. However the traditional method of inferring these values manually is such an onerous task that automated methods of cell tracking and segmentation are in high demand. Image cytometry is a practical approach to measure and extract cell properties from large volumes of microscopic cell images. As an important application of image cytometry, this paper presents a probabilistic model based cell tracking method to locate and associate HSCs in phase contrast microscopic images. The proposed cell tracker has been successfully applied to track HSCs based on the most probable identified cell locations and probabilistic data association.

In this paper, we propose a new scheme for extracting the contour of the carotid artery using ultrasound images. Starting from a user defined seed point within the artery, the scheme uses the fuzzy region growing algorithm to create a fuzzy connectedness map for the image. Then, the fuzzy connectedness map is thresholded using a threshold selection mechanism to segment the area inside the artery. Experimental results demonstrated the efficiency of the proposed scheme in segmenting carotid artery ultrasound images, and it is insensitive to the seed point location, as long as it is located inside the artery.

This paper presents a new cDNA microarray image segmentation framework. The framework uses robust vector median filtering to generate a root sigLnal which is an image obtained from the input by repeatedly filtering it until no more changes occur. During the convergence to the root signal, the framework classifies the cDNA image data as either microarray spots or image background, and ideally separates the regular spots from the background. Thus, the obtained root signal represents the segmented microarray image. In addition, the framework excellently removes noise present in the cDNA microarray images and normalizes spots’ intensities.

One of the key issues in microarray analysis is to extract quantitative information from the spots, which represents gene expression levels in the experiments. The process of identifying the spots and separating the foreground from the background is known as microarray image segmentation. In this paper, we propose a new approach to microarray image segmentation, which we called the , and shows various advantages when compared to the adaptive circle method. Our experiments on real-life microarray images show that adaptive ellipse is capable of extracting information from the images, which is ignored by the traditional adaptive circle method, and hence showing more flexibility.

Retinal exudates are typically manifested as spatially random yellow/white patches of varying sizes and shapes. They are a visible sign of retinal diseases such as diabetic retinopathy. Following some key preprocessing steps, colour retinal image pixels are classified to exudate and non-exudate classes. nearest neighbour, Gaussian quadratic and Gaussian mixture model classifiers are investigated within the pixel-level exudate recognition framework. A Gaussian mixture model-based classifier demonstrated the best classification performance with 89.2% and 81.0% in terms of pixel-level accuracy and 92.5% and 81.4% in terms of image-based accuracy.

This paper presents two novel feature selection techniques for the purpose of prostate tissue characterization based on Trans-rectal Ultrasound (TRUS) images. First, suspected cancerous regions of interest (ROIs) are identified from the segmented TRUS images using Gabor filters. Next, second and higher order statistical texture features are constructed for these ROIs. Furthermore, a representative feature subset with the best discriminatory power among the constructed features is selected using two artificial life techniques: the Particle Swarm Optimization (PSO) and the Ant Colony Optimization (ACO). Both the PSO and ACO are tailored to fit the binary nature of the feature selection problem. The results are compared to the results obtained using the Genetic Algorithm (GA) feature selection approach. When Support Vector Machine (SVM) classifier is applied for the purpose of tissue characterization, the features obtained using the PSO and ACO outperforms the features obtained using the GA, i.e., they are capable of discriminating between suspicious cancerous and non-cancerous in a better accuracy. The obtained results demonstrate excellent tissue characterization with 83.3% sensitivity, 100% specificity and 94% overall accuracy.