Catálogo de publicaciones - libros

A method for visualization the correlation-based data has been investigated. The advantage of this method lies in the possibility to restore the system of multidimensional vectors describing parameters from their correlation matrix (one vector for one parameter) and to visualise these vectors for the visual decision making on the similarity of the parameters. The goal of this research is to investigate the possibility to reduce the dimensionality of the vectors from the restored system and to evaluate the visualization quality in dependence on the reduction level.

We present a method for image segmentation, that is, to identify image points with an indication of the region or class they belong to. The proposed algorithm basically consists of two stages. First it starts by restoring the image from possible contamination. In the second stage it analyzes each pixel using a 3x3 sliding window. For the first pixel, it creates an objets consisting of that same pixel, and registers this object in an array. In the subsequent steps, a cluster analysis is applied to the surrounding eight pixels, an determines whether the central pixel belongs to one of the existing objects, or a new object has to be created, and registered in the array of objects.

Image metamorphosis, commonly known as morphing, is a powerful tool for visual effects that consists of the fluid transformation of one digital image into another. There are many techniques for image metamorphosis, but in all of them there is a need for a person to supply the correspondence between the features in the source image and target image. In this paper we use the Active Shape Models and Evolution Strategies to perform the metamorphosis of face images in frontal view automatically.

Generally, it is difficult to find constant patterns on identifiers in a container image, since the identifiers are not normalized in color, size, and position, etc. and their shapes are damaged by external environmental factors. This paper distinguishes identifier areas from background noises and removes noises by using an ART2-based quantization method and general morphological information on the identifiers such as color, size, ratio of height to width, and a distance from other identifiers. Individual identifier is extracted by applying the 8-directional contour tracking method to each identifier area. This paper proposes a refined ART2-based RBF network and applies it to the recognition of identifiers. Through experiments with 300 container images, the proposed algorithm showed more improved accuracy of recognizing container identifiers than the others proposed previously, in spite of using shorter training time.

Based on a multilayer perceptron (MLP), a blind image restoration method is presented. The algorithm considers both local region information and edge information of an image. To reduce the dimension of the network’s input, a sliding window approach is employed to extract the features of the blurred image, which makes use of local region information. For the purpose of accelerating training and improving the restoration performance, the edge part and the smooth part in an image are separated and then used as training sets, respectively. A mapping model between the blurred image and the clear one is established through training the MLP with LM algorithm and then it is utilized to restore the blurred image. The simulation results demonstrate the proposed method feasible for image restoration.

Reflective symmetry is useful for various areas such as computer vision, medical imaging, and 3D model retrieval system. This paper presents an intuitive reflective symmetry detection method for 3D polygon objects. Without any mapping process the method detects the reflective symmetry plane by parallel projection. This paper defines a continuous measure to estimate how much an object is reflective symmetrical for a projection plane through the center of the object. Also it explores the method to detect the reflective symmetry plane with the measure. The proposed method can detect up to 99% reflective symmetry plane not exceeding 4 degree angle for perfect symmetry objects and detect up to 85% reflective symmetry plane not exceeding 10 degree angle for near symmetry objects using Princeton Shape Benchmark.

This paper proposes a new filtering scheme for eliminating random-valued impulse noise from gray images. In the first phase a noise detector is utilized to extract the noise candidates. Next, the algorithm applies a connected component analysis in order to gather the neighboring noisy pixels into separate sets of connected noise candidates. The corrupted pixels are restored using a detail preserving regularization method. The main idea of the proposed approach is to gather the noisy candidate pixels into separate sets of connected pixels and solve the minimization functional over these pixels. Experimental results illustrate the efficiency and effectiveness of the algorithm.

In this paper an efficient queue-based algorithm for order independent homotopic thinning is proposed. This generic algorithm can be applied to various thinning versions: homotopic marking, anchored skeletonisation, and the computation of the skeleton of influence zones based on local pixel characterisations. An example application of the proposed method to detect the medial axis of wide river networks from satellite imagery is also presented.

Segmentation is a fundamental task in pattern recognition and basis for high level applications like scene reconstruction, change detection, or object classification. The performance of these tasks suffers from a distorted segmentation. In this contribution an adaptive diffusion-based segmentation method is proposed suppressing perturbations in the segmentation with focusing on small regions with high contrast to their surrounding. The algorithm determines in each step the diffusion tensor. It is re-weighted with respect to an assessment stage. A comparative study uses high-resolution remote sensing data.

In this paper we propose a method of using Weighted Order Statistic (WOS) filters for the task of pattern detection. Usually WOS filters are applied to noise removal. An efficient algorithm for pattern detection is described in details with emphasis put on the problem of a proper choice of filter windows. Also practical results of different pattern detection cases are presented.