Catálogo de publicaciones - libros

A morphological edge detector for robust real time image segmentation is proposed in this paper. Different from traditional thresholding methods that determine the threshold based on image gray level distribution, our method derives the threshold from object boundary point gray values and the boundary points are detected in the image using the proposed morphological edge detector. Firstly, the morphological edge detector is applied to compute the image morphological gradients. Then from the resultant image morphological gradient histogram, the object boundary points can be selected, which have higher gradient values than those of points within the object and background. The threshold is finally determined from the object boundary point gray values. Thus noise points inside the object and background are avoided in threshold computation. Experimental results on currency image segmentation for real time printing quality inspection are rather encouraging.

In this paper we extend the basic morphological operators dilation and erosion for grey-scale images based on the threshold approach, umbra approach and fuzzy set theory to colour images. This is realised by treating colours as vectors and defining a new vector ordering so that new colour morphological operators are presented. Here we only discuss colours represented in the RGB colour space. The colour space RGB becomes together with the new ordering and associated minimum and maximum operators a complete chain. All this can be extended to the colour spaces HSV and L*a*b*. Experimental results show that our method provides an improvement on the component-based approach of morphological operators applied to colour images. The colours in the colour images are preserved, that is, no new colours are introduced.

Morphological operations with 3D images require a huge amount of computation. The decomposition of structuring elements used in the morphological operations such as dilation and erosion greatly reduces the amount of computation. This paper presents a new method for the decomposition of a 3D convex structuring element into a set of basis convex structuring elements. Furthermore, the decomposition method is applied to the neighborhood decomposition, in which each basis is one of the combinations of the origin voxel and its 26 neighborhood voxels. First, we derived the set of decomposition conditions on the lengths of the original and the basis convex structuring elements, and then the decomposition problem is converted into a linear integer optimization problem. The objective of the optimization is to minimize a cost function representing the optimal criterion of the parallel processing computer architecture on which the operation is performed. Thus, our method can be used to obtain the different optimal decompositions minimizing the amount of computation for different parallel processing computer architectures.

In this paper a novel class of filters designed for the removal of impulsive noise in color images is presented. The proposed filter family is based on the kernel function which regulates the noise suppression properties of the proposed filtering scheme. The comparison of the new filtering method with standard techniques used for impulsive noise removal indicates superior noise removal capabilities and excellent structure preserving properties.

A method for color indexing is proposed that is based upon nonparametric statistical techniques. Nonparametrics compare the ordinal rankings of sample populations, and maintain their significance when the underlying populations are not Normally distributed. The method differs from previous approaches to color indexing, in that it does not involve histogramming. Principal component analysis is performed to extract the three orthogonal axes of maximum dispersion for a given color signature. These axes are then used to select Lipschitz embeddings to generate sets of scalars that combine all color channel information. These scalar sets are compared against a ranked database of such scalars using the Moses test for variance. On the resulting top matches, the Wilcoxon test of central tendency is applied to yield the best overall match.

The method has been tested extensively on a number of image databases, and has been compared against eight standard histogram methods using four color space transformations. The tests have shown its performance to be competitive with, and in certain cases superior to, the best histogram methods. The technique also shows a greater robustness to noise than all histogram methods, with a noise robustness comparable to that of the more expensive Variable Kernel Density method.

A novel noniterative extrapolation method based on Taylor series approximation is proposed for color filter array demosaicing. New extrapolation equations are derived for the estimation of the green plane with higher accuracy by including higher order terms of the Taylor series. Our proposed method avoids interpolation across an edge and thus reduces errors in the demosaiced image. It has been shown that our method outperforms other techniques in image quality measures, especially around edges.

A camera is an effective tool in capturing images for colorimetric use. However, the signals generated by different cameras are not equal for the same scene. Therefore, cameras are characterized based on a CIE standard colorimetric observer. This paper proposes a new method for obtaining camera transfer matrices under different white balances using a 3×3 camera transfer matrix under a specific white balance point. As such, the proposed methods enables a camera transfer matrix under any other white balance to be obtained using the colorimetric coordinates for the phosphor primaries derived from a 3×3 linear transfer matrix under a certain white balance point. Experimental results confirmed that the proposed method produced a 3×3 linear transfer matrix under any other white balance with a reasonable degree of accuracy compared with the transfer matrix obtained by the conventional method.

Color is an important cue both in machine vision and image processing applications, despite its dependence upon illumination changes. We propose a color constancy algorithm that estimates both the set and the likelihood of feasible color mappings in respect to their frequency and effectiveness. The best among this set is selected to rendered back image colors as seen under a canonical light. Experiments were done to evaluate its performance compared to Finlayson’s 2D gamut–mapping algorithm, outperforming it. Our approach is a helpful alternative wherever illumination is poorly known since it employs only image data.

In this paper, we report the results of a comparative study on skin-color models generally used for facial region location. These include two 2D Gaussian models developed in normalized RGB and HSV color spaces respectively, a 1D lookup table model of hue histogram, and an adaptive 3D threshold box model. Also, we present a new model – called “adaptive hue lookup table”. The model is developed by introducing the so-called “Continuously Adaptive Mean Shift” (Camshift) technique into a traditional hue lookup table method. With the introduction of Camshift, the lookup table is able to adaptively adjust its parameters to fit the illumination conditions of different test images. In the experiments reported here, we compare the proposed method with the four typical skin-color filters in the scenarios of different human races and illuminations. The obtained results indicate that the proposed method reaches the best balance between false detection and detect rate.

We present a texture analysis approach for texture image indexing based on Gabor-like Hermite filters, which are steered versions of discrete Hermite filters. Hermite filters are the backbone of the Hermite transform, which is a polynomial transform and a good model of the human visual system. Experimental results show that our filters have better performance than Gabor filters. The texture analysis system is then applied to handwriting document indexing. For that doing, handwriting documents are decomposed into local frequencies through the presented filter bank and, using this decomposition, we analyze the visual aspect of handwritings to compute similarity measures. A direct application is the management of document databases, allowing to find documents coming from the same author or to classify documents containing handwritings that have similar visual aspect. The current results are very promising and show that it is possible to characterize handwritten drawings without any a priori graphemes segmentation.