Catálogo de publicaciones - libros

In this work the relation between scale-space image segmentation and selection of the localization scale is examined first, and a scale selection approach is consequently proposed in the segmentation context. Considering the segmentation part, gradient watersheds are applied to the non-linear scale-space domain followed by a grouping operation. A report on localization scale selection techniques is carried out next. Furthermore a scale selection method that originates from the evolution of the probability distribution of a region uniformity measure through the generated scales is proposed. The introduced algorithm is finally compared to a previously published approach that is also introduced into the segmentation context to indicate its efficacy.

Colour prediction models (CPM) can be used to analyze the printing quality of halftone-based color printing systems. In this paper, we consider the Neugebauer CPM which requires as input the fraction of occupation of each primary. To obtain these numbers, we apply several image segmentation algorithms, with and without contextual information. These segmentation algorithms are evaluated with respect to another technique based on mixtures of factor analyzers. More importantly, the segmentation results are evaluated with respect to the performance of the Neugebauer CPM when used with the obtained fractions of occupation. This evaluation is carried out by comparing the predicted color against that measured with a spectrophotometer, and testifies for the adequacy of the approach.

A modified FCM algorithm based on spatial and multiresolution constraints is described in this paper. First the pyramid is built from the original image then in each level FCM parameters are computed under a neighborhood spatial constraint. The coarse membership functions propagate down to fine layers to improve segmentation accuracy. The algorithm is tested on both synthetic and multispectral images. Experimental results are presented, showing the effectiveness of the method.

A novel segmentation algorithm for natural color image is proposed. Fibonacci Lattice-based Sampling is used to get the color labels of image so as to take advantage of the traditional approaches developed for gray-scale images. Using local fuzzy homogeneity derived from color labels, texture component is calculated to characterize spatial information. Color component is obtained by peer group filtering. To avoid over-segmentation of texture areas in a color image, these color and texture components are jointly employed to group the pixels into homogenous regions by the mean shift based clustering. Finally, experiments show very promising results.

A novel method of unsupervised imagesegmentation using contourlet domain hidden markov trees model is presented. Fuzzy C-mean clustering algorithm is used to capture the likelihood disparity of different texture features. A new context based fusion model is given for preserve more interscale information in contourlet domain. The simulation results of synthetic mosaics and real images show that the proposed unsupervised segmentation algorithm represents a better performance in edge detection and protection and its error probability of the synthetic mosaics is lower than wavelet domain HMT based method.

C-V method, an active contour model developed by Chan and Vese, has been successfully applied to solve the problem of object detection in gray-scale images. In this paper, a novel color C-V method which takes into account of color information and global property is presented. Choosing the appropriate color space for this model is also introduced. Finally, the applications of the proposed method to natural color images and microscopic halftone printing images are given and the experimental results show robust performance especially in case of weak edges and noisy inputs.

The presence of speckle in synthetic aperture radar (SAR) images makes the segmentation of such images difficult. In this paper, a set of energy measures of channels of the undecimated wavelet decomposition is used to represent the texture information of SAR image efficiently. Furthermore, the kernel FCM incorporating spatial constraints, which is characteristic of robustness to noise, is applied to the SAR image segmentation. A synthesized texture image and a Ku-band SAR image are used in experiments and the successful segmentation results show the validation of the method.

This paper addresses the segmentation of crystalline Zinc oxide nanocolumns from microscopic images. ZnO is a direct band semiconductor suitable for many applications whose interest has been growing recently. One of these applications are light-collecting devices such as solar cells, using nanostructured substrates. Electrodeposition is a low cost technique very suitable for the preparation of nanostructured ZnO, producing nanocolumnar ZnO crystals with a morphology that depends on the deposition parameters and the substrate characteristics. The parameters of the sample can be determined processing images of the nanostructures, which is the objective of this study.

Current methods of multimodal image registration usually seek to maximize the similarity measure of mutual information (MI) between two images over their region of overlap. In applications such as planned radiation therapy, a diagnostician is more concerned with registration over specific regions of interest (ROI) than registration of the global image space. Registration of the ROI can be unreliable because the typically small regions have limited statistics and thus poor estimates of entropies. We examine methods to improve ROI-based registration by using information from the global image space.

We combine the techniques of the complex wavelet transform and Markov random fields (MRF) model to restore natural images in white Gaussian noise. The complex wavelet transform outperforms the standard real wavelet transform in the sense of shift-invariance, directionality and complexity. The prior MRF model is used to exploit the clustering property of the wavelet transform, which can effectively remove annoying pointlike artifacts associated with standard wavelet denoising methods. Our experimental results significantly outperform those using standard wavelet transforms and are comparable to those from overcomplete wavelet transforms and MRFs, but with much less complexity.