Catálogo de publicaciones - libros

Local alignment of two protein sequences shows the similar regions between these proteins. Usually, a query protein sequence is aligned with several hundred thousands of protein sequences stored in databases. Since this procedure is computationally demanding, various hardware units are designed to get high quality results in a practically useful time. This paper presents efficient hardware designs that compute the local alignment scores of protein sequences. The presented designs are compared with the reference designs. All designs are implemented using ASIC and FPGA technologies. Syntheses results show that compared to the reference designs the proposed ASIC implementations achieve frequency improvements up to 250 % and hardware gains up to 40 %, and the proposed FPGA implementations achieve frequency improvements up to 29 % and hardware gains up to 48 %.

Most models of Bidirectional associative memories intend to achieve that all trained pattern correspond to stable states; however, this has not been possible. Also, none of the former models has been able to recall all the trained patterns. In this work we introduce a new model of bidirectional associative memory which is not iterative and has no stability problems. It is based on the Alpha-Beta associative memories. This model allows, besides correct recall of noisy patterns, perfect recall of all trained patterns, with no ambiguity and no conditions. An example of fingerprint recognition is presented.

We propose a new data structure to accelerate the color computation step of CPU-based volume ray casting. To ensure interactive frame rates on a PC platform, we store interpolated scalar value and gradient vector required for color computation step in volume ray casting. However, it is difficult to store those two values in preprocessing step because sample points can lie in arbitrary position in a cell. Therefore, after determining candidate cells that contribute to the final images, we partition each candidate cell into several sub-cells. Then, we store trilinearly interpolated scalar value and an index of encoded gradient vector for each sub-cell. Because the information that requires time-consuming computations is already stored in our data structure, color values are determined without further computations.

The paper proposes some major modifications on the mass-spring chain algorithm. The original algorithm takes into account individual springs of the 3 D lattice representing a deformable object and deforms springs. In this proposed version, deformation algorithm loops through triangles of the 3 D mesh and performs deformation. Two additional improvements are also provided: A new formula, which produces smooth deformations, is used for deformation calculation. A simple and fast algorithm for crossing test in order to avoid vertex penetration is also employed. Modified mass-spring chain algorithm is used in several applications and is compared against the original algorithm. The proposed method is faster than the original algorithm and suitable for deformation simulations in virtual realty applications where real time performance is required.

Quadtree-based terrain visualization methods have been used in a lot of applications. However, because most procedures are performed on the CPU, the rendering speed is slow. In this paper, we present a quadtree-based terrain visualization method working on the GPU with specially designed data structure, the error-buffer and flag-buffer named ef -buffers. In pre-processing step, error metrics are computed in world space and the error metrics are transferred to the error-buffer. In rendering time, LOD selection and view-frustum culling are processed by evaluating the error metrics. The result is stored into the flag-buffer. To remove cracks or T-junction, the flag-buffer is refined. Then triangulation is performed using the flag-buffer. This method reduces CPU load and performs time consuming jobs such as LOD selection and view-frustum culling on the GPU. We can conclude that our method much faster than CPU-based rendering method without loss of image quality.

In this work, the appearance of woven fabric defects due to yarn faults has been simulated by a method developed which is based on photographs taken from yarns along their lengths. Simulation of woven fabric is based on yarn and fabric cross sectional geometry as well as on yarn measurements and images. The transformation of yarn from a straight circular cylinder to a flattened and crimped form in fabric structure is modelled mathematically and this is applied to obtain its projected image on the fabric surface by resizing the yarn image on the photographic plane. The resized yarn images thus obtained are, then, rearranged to simulate pixel based fabric surface appearance onto which images of faulty yarn sections are superposed. This is achieved by a suitable software developed, based on raster graphics. The faulty fabric simulations are confined to single fabrics of plain, matt and twill weaves with neps, slubs, thick and thin places.

A method using a ray-coherence in volume ray casting skips over transparent region efficiently because a current ray advances as the amount of the leaping distance between an image plane and the object boundary of a representative ray. If the ray jumps over the object boundary, a user-selected constant distance of which the size is larger than the unit distance is commonly used to reach the boundary quickly. However, since we cannot determine an accurate distance, the traversal speed depends on the constant distance. In this paper, to support efficient space-leaping in transparent region and to determine the distance from inside of an object to its boundary, we generate a distancemap for transparent region as well as for nontransparent region, named bidirectional distancemap. In rendering step, a current ray advances as the amount of the leaping distance of a representative ray. If it lies on transparent region, we use our distancemap for transparent region such as the previous approaches. Otherwise, when it already jumps over an object boundary, the distancemap for nontransparent region is exploited.

Segmenting 3D object surfaces is required for various high level computer vision and computer graphics applications. In computer vision, recognizing and estimating poses of 3D objects heavily depend on segmentation results. Similarly, physically meaningful segments of a 3D object may be useful in various computer graphics applications. Therefore, there are many segmentation algorithms proposed in the literature. Unfortunately, most of these algorithms can not perform reliably on free-form objects. In order to segment free-form objects, we introduce a novel method in this study. Different from previous segmentation methods, we first obtain a function representation of the object surface in spherical coordinates. This representation allows detecting smooth edges on the object surface easily by a zero crossing edge detector. Edge detection results lead to segments of the object. We test our method on diverse free-form 3D objects and provide the segmentation results.

In this paper, we use the natural neighbor coordinates and its properties as a tool for electromagnetic tracker calibration. We have established a framework for computing the three-dimensional natural neighbor coordinates and used it for the calibration process for which we wish to eliminate the distortion error presented in the electromagnetic field. Our work comprises of two main parts. The first is about the computational model of the natural neighbor coordinates while the second is a discussion on the result we get after calibrating the electromagnetic tracker distortion. Our result has shown that the natural neighbor coordinates can be a promising method for approximating the distortion in the electromagnetic field and for calibrating the distorted field data.

This paper presents an easy-to-implement and efficient method to calculate the Minkowski Sums of simple convex objects. The method is based on direct manipulation of planes in 3D space. The paper then explains how this method is used for generating octrees for scenes consisting of triangular meshes.