Catálogo de publicaciones - libros

In this work we describe a novel statistical video representation and modeling scheme. Video representation schemes are needed to enable segmenting a video stream into meaningful video-objects, useful for later indexing and retrieval applications. In the proposed methodology, unsupervised clustering via Guassian mixture modeling extracts coherent space-time regions in feature space, and corresponding coherent segments () in the video content. A key feature of the system is the analysis of video input as a single entity as opposed to a sequence of separate frames. Space and time are treated uniformly. The extracted space-time regions allow for the detection and recognition of video events. Results of segmenting video content into static vs. dynamic video regions and video content editing are presented.

In this paper we describe a flexible approach for determining the relative orientation of the camera with respect to the scene. The main premise of the approach is the fact that in man-made environments, the majority of lines is aligned with the principal orthogonal directions of the world coordinate frame. We exploit this observation towards efficient detection and estimation of vanishing points, which provide strong constraints on camera parameters and relative orientation of the camera with respect to the scene.

By combining efficient image processing techniques in the line detection and initialization stage we demonstrate that simultaneous grouping and estimation of vanishing directions can be achieved in the absence of internal parameters of the camera. Constraints between vanishing points are then used for partial calibration and relative rotation estimation. The algorithm has been tested in a variety of indoors and outdoors scenes and its efficiency and automation makes it amenable for implementation on robotic platforms.

The content-plot of a video clip is created by positioning several key frames in two-dimensions and connecting them with lines. It is constructed so that it should be possible to follow the events shown in the video by moving along the lines. Content plots were previously computed by clustering together frames that are contiguous in time. We propose to cluster together frames if they are related by a short chain of similarly looking frames even if they are not adjacent on the time-line. The computational problem can be formulated as a graph clustering problem that we solve by extending the classic k-means technique to graphs. This new graph clustering algorithm is the main technical contribution of this paper.

Retinal exudates are a characteristic feature of many retinal diseases such as Diabetic Retinopathy. We address the development of a method to quantitatively diagnose these random yellow patches in colour retinal images automatically. After a colour normalisation and contrast enhancement preprocessing step, the colour retinal image is segmented using Fuzzy C-Means clustering. We then classify the segmented regions into two disjoint classes, exudates and non-exudates, comparing the performance of various classifiers. We also locate the optic disk both to remove it as a candidate region and to measure its boundaries accurately since it is a significant landmark feature for ophthalmologists. Three different approaches are reported for optic disk localisation based on template matching, least squares are estimation and snakes. The system could achieve an overall diagnostic accuracy of 90.1% for identification of the exudate pathologies and 90.7% for optic disk localisation.

Active shape models (ASMs) have been shown to be a powerful tool to aid the interpretation of images. ASM model parameter estimation is based on the assumption that residuals between model fit and data have a Gaussian distribution. However, in many real applications, specifically those found in the area of medical image analysis, this assumption may be inaccurate. Robust parameter estimation methods have been used elsewhere in machine vision and provide a promising method of improving ASM search performance. This paper formulates M-estimator and random sampling approaches to robust parameter estimation in the context of ASM search. These methods have been applied to several sets of medical images where ASM search robustness problems have previously been encountered. Robust parameter estimation is shown to increase tolerance to outliers, which can lead to improved search robustness and accuracy.

In this paper, a new image-matching mathematical model is presented for the mammogram registration. In avariational framework, an energy minimization problem is formulated and a multigrid resolution algorithm is designed. The model focuses on the matching of regions of interest. It also combines several constraints which are both intensity and segmentation based. A new feature of our model is combining region matching and segmentation by formulation of the energy minimization problem with free boundary conditions. Moreover, the energy has a new registration constraint. The performances of models with and without free boundary are compared on a simulated mammogram pair. It is shown that the new model with free boundary is more robust to initialization inaccuracies than the one without. The interest of the new model for the real mammogram registration is also illustrated.

Image segmentation is a fundamental problem in Image Processing, Computer Vision and Medical Imaging with numerous applications. In this paper, we address the atlas-based image segmentation problem which involves registration of the atlas to the subject or target image in order to achieve the segmentation of the target image. Thus, the target image is segmented with the assistance of a registration process. We present a novel variational formulation of this registration assisted image segmentation problem which leads to solving a coupled set of nonlinear PDEs that are solved using efficient numerical schemes. Our work is a departure from earlier methods in that we have a wherein registration and segmentation are simultaneously achieved. We present several 2D examples on synthetic and real data sets along with quantitative accuracy estimates of the registration.

Automatic 3D segmentation of the brain from MR scans is a challenging problem that has received enormous amount of attention lately. Of the techniques reported in literature, very few are fully automatic. In this paper, we present an efficient and accurate, fully automatic 3D segmentation procedure for brain MR scans. It has several salient features namely, (1) instead of a single multiplicative bias field that affects all tissue intensities, separate parametric smooth models are used for the intensity of each class. This may be a more realistic model and avoids the need for a logarithmic transformation. (2) A brain atlas is used in conjunction with a robust registration procedure to find a non-rigid transformation that maps the standard brain to the specimen to be segmented. This transformation is then used to: segment the brain from non-brain tissue; compute prior probabilities for each class at each voxel location and find an appropriate automatic initialization. (3) Finally, a novel algorithm is presented which is a variant of the EM procedure, that incorporates a fast and accurate way to find optimal segmentations, given the intensity models along with the spatial coherence assumption. Experimental results with both synthetic and real data are included, as well as comparisons of the performance of our algorithm with that of other published methods.

A novel approach for computing point correspondences and grids within annular tissues is presented based on a recently introduced technique for computing thickness in such regions. The solution of Laplace’s equation provides implicit correspondence trajectories between the bounding surfaces. Pairs of partial differential equations are then efficiently solved within an Eulerian framework for thickness, from which concentric surfaces can be constructed. Point correspondences are then computed between the outer surfaces and any surface within, providing a gridding of the annular tissue. Examples are shown for two-dimensional short-axis images of the left ventricle and three-dimensional images of the cortex.

Detection followed by morphological processing is commonly used in machine vision. However, choosing the morphological operators and parameters is often done in a heuristic manner since a statistical characterization of their performance is not easily derivable. If we consider a morphology operator sequence as a classifier distinguishing between two patterns, the automatic choice of the operator sequence and parameters is possible if one derives the misclassification distribution as a function of the input signal distributions, the operator sequence, and parameter choices. The main essence of this paper is the illustration that misclassification statistics, the distribution of bit errors measured by the Hamming distance, can be computed by using an embeddable Markov chain approach. License plate extraction is used as a case study to illustrate the utility of the theory on real data.