Catálogo de publicaciones - libros

This paper proposes a new method to create realistic composition images using the shadow extracted from the real images for the virtual studio. To represent the shadow of human on the floor of the virtual space, the real image is separated the human and shadow, and the shadow matched to the color of the virtual space is composed. Next, to represent the shadow on the virtual object, the deformed shadow is projected on the virtual object. The results show the effectiveness of the proposed method to represent the shadow for the virtual studio.

The growing heterogeneity of networks, terminals and users and the increasing availability and usage of multimedia content have been raising the relevance of content adaptation technologies able to fulfill the needs associated to all usage conditions without multiplying the number of versions available for the same piece of content. This paper discusses the problem of content adaptation considering the major technologies which may have a role in achieving the ‚one fits all’ content provision paradigm.

Center-point model of deformable surface for segmentation of 3D images is presented. Mobility of each node, element composing the surface, is constrained in the model to one direction. Also an original formula for image influence computation is proposed. The model aims at algorithm simplification and reduction of computational time needed for segmentation of 3D imaging data acquired from magnetic resonance or computer tomography scanner.

An essential step in the analysis of gene expression profile data is the detection of gene groups that have similar expression patterns. Although many clustering algorithms have been proposed for such task, problems such as visualizing the clustering results are still not satisfactorily addressed. In this paper, a novel methodology for drawing the gene clusters in 3D is proposed. The algorithm firstly allocates the genes within a cluster to a local area – InfoCube using Force-Directed Placement Spring Model; it then allocates all the InfoCubes within a global area using the same method. The bottom-up approach saves time in coordinates’ computation and successfully avoids the space partition problem in multi-layer graph drawing. It is not only effective in displaying the double-layer clustering results but also can be extended to display other multi-layer graphs with hierarchical relationships.

The technique of independent component analysis (ICA) is applied for texture feature detection. In ICA an optimal transformation (with respect to the statistical structure of the image samples set) is discovered via blind signal processing. Any texture is considered as a mixture of independent sources (basic functions of detected transformation). Experimental comparison is documented on the compactness and separability of base functions, the data-specific ICA-based and universal Gabor functions (the latter are set by default for all kinds of images).

The evidence accumulation method for finding objects having shape which can be neither parameterized nor tabularized is proposed. The result is a multi-scale measure of existence of the detected object, in the accumulator congruent with the image domain, supplemented with local information on additional features of the object. The method is implemented for finding blood vessels in mammographic images, visible as bright lines. In this case, information from pairs of pixels is used for accumulation. The accumulation is fuzzy in several ways.

.

Facial feature extraction is necessary for identification of an individual face on a computer. We propose an automatic facial feature extraction method that is robust in conditions of an input image. As facial features, the shape of facial parts is automatically extracted from a frontal face image. We experiment with input images of several resolutions that are taken with various types of camera. The accuracy of the extraction is shown by comparing our results with the corresponding manually extracted facial features.

Several algorithms of intensity-based feature detection have been recently proposed. The paper further investigates one of them. The algorithm employs locally computed moments to detect image features. First, the best-match template feature is built for the current location. Then, the template is compared to the content of the image to determine the actual presence of the feature. The paper reports an improvement in the second step. The template feature and the image are compared using radial profiles. It improves the performance (especially for noised and textured images) and simplifies the prospective hardware implementation of the algorithm.

The paper addresses the problem of multi-spectral image segmentation based on texture detection in domain of scale-space. The main contribution of this paper is presentation of the robust image partitioning method based on the tensor operator for detection of local structures in neighborhoods of pixels. We also extend the concept of a structural tensor to multi- spectral images and discuss the two different concepts of a scale associated with this tensor. The presented method was tested with many different monochrome and color images. We provide the experimental results and details of implementation.