Catálogo de publicaciones - libros

We present in this paper a new image scaling algorithm which is based on the generalized sampling theorem of Papoulis. The main idea consists in using the first and second derivatives of an image in the scaling process. The derivatives contain information about edges and discontinuities that should be preserved during resizing. The sampling theorem of Papoulis is used to combine this information. We compare our algorithm with eight of the most common scaling algorithms and two measures of quality are used: the standard deviation for evaluation of the blur, and the curvature for evaluation of the aliasing. The results presented here show that our algorithm gives the best images with very few aliasing, good contrast, good edge preserving and few blur.

In this paper we present a new alternative noise reduction method for color images that were corrupted with additive Gaussian noise. We illustrate that color images have to be processed in a different way than most of the state-of-the-art methods. The proposed method consists of two sub-filters. The main concern of the first subfilter is to distinguish between local variations due to noise and local variations due to image structures such as edges. This is realized by using the color component distances instead of component differences as done by most current filters. The second subfilter is used as a complementary filter which especially preserves differences between the color components. This is realized by calculating the local differences in the red, green and blue environment separately. These differences are then combined to calculate the local estimation of the central pixel. Experimental results show the feasibility of the proposed approach.

In imaging applications such as ultrasound or Synthetic Aperture Radar (SAR), the local texture features are descriptive of the types of geological formation or biological tissue at that spatial location. Therefore, on denoising these texture images, it is essential that the local texture details characterizing the geological formation or tissue type are not lost. When processing these images, the operator usually has prior knowledge of the type of textures to expect in the image. In this work, this prior knowledge is exploited to implement a spatially-adaptive and subband-adaptive wavelet threshold that denoises texture images while preserving the characteristic features in the textures. The proposed algorithm involves three stages: texture characterization, texture region identification system training, and texture region identification and denoising. In the first stage, the texture features to be preserved are characterized by the subband energies of the wavelet decomposition details at each level. Next, the energies of the characteristic subband are used as inputs to train the adaptive neural-fuzzy inference system (ANFIS) classifier for texture region identification. Finally, the texture regions are identified by the ANFIS and the subband-adaptive BayesShrink threshold is adjusted locally to obtain the proposed spatially-adaptive and subband-adaptive threshold.

A new image-denoising algorithm is proposed, which uses the Generalized Cross Validation technique in the ridgelet frame domain. The proposed algorithm has two advantages: first, it can select the optimal threshold automatically without the knowledge of the noise level; second, it has the ability to well recover the ‘line-type’ structures contained in noisy images. Experimentally, the high performance of the proposed algorithm is demonstrated.

A discrete regularization framework on graphs is proposed and studied for color image filtering purposes when images are represented by grid graphs. Image filtering is considered as a variational problem which consists in minimizing an appropriate energy function. In this paper, we propose a general discrete regularization framework defined on weighted graphs which can be seen as a discrete analogue of classical regularization theory. With this formulation, we propose a family of fast and simple anisotropic linear and nonlinear filters. The parameters of the proposed discrete regularization are estimated to have a parameterless filtering.

In this paper, we study the problem of restoring multicomponent images. In particular, we investigate the effects of accounting for the correlation between the image components on the deconvolution and denoising steps. The proposed restoration is a 2-step procedure, comprising a shrinkage in the Fourier domain, followed by a shrinkage in the wavelet domain. The Fourier shrinkage is performed in a decorrelated space, by performing PCA before the Fourier transform. The wavelet shrinkage is performed in a Bayesian denoising framework by applying multicomponent probability density models for the wavelet coefficients that fully account for the intercomponent correlations. In an experimental section, we compare this procedure with the single-component analogies, i.e. performing the Fourier shrinkage in the correlated space and using single-component probability density models for the wavelet coefficients. In this way, the effect of the multicomponent procedures on the deconvolution and denoising performance is studied experimentally.

This paper introduces a Bayesian restoration method for low-resolution images combined with a smoothness prior and a newly proposed adaptive bimodal prior. The bimodal prior is based on the fact that an edge pixel has a colour value that is typically a mixture of the colours of two connected regions, each having a dominant colour distribution. Local adaptation of the parameters of the bimodal prior is made to handle neighbourhoods which have typically more than two dominant colours. The maximum a posteriori estimator is worked out to solve the problem. Experimental results confirm the effectiveness of the proposed bimodal prior and show the visual superiority of our reconstruction scheme to other traditional interpolation and reconstruction methods for images with a strong colour modality like cartoons: noise and compression artefacts are removed very well and our method produces less blur and other annoying artefacts.

In this paper, we approach the problem of determining the basic components from repetitive textured surfaces undergoing free-form deformations. Traditional methods for texture modeling are usually based on measurements performed on fronto-parallel planar surfaces. Recently, affine invariant descriptors have been proposed as an effective way to extract local information from non-planar texture surfaces. However, affine transformations are unable to model local image distortions caused by changes in surface curvature. Here, we propose a method for selecting the most representative candidates for the basic texture elements of a texture field while preserving the descriptors’ affine invariance requirement. Our contribution in this paper is twofold. First, we investigate the distribution of extracted affine invariant descriptors on a nonlinear manifold embedding. Secondly, we describe a learning procedure that allows us to group repetitive texture elements while removing candidates presenting high levels of curvature-induced distortion. We demonstrate the effectiveness of our method on a set of images obtained from man-made texture surfaces undergoing a range of non-rigid deformations.

A common issue in many computer vision applications is the effect of the illumination conditions on the performance and reliability of the built system. In many cases the researchers have to face an extra problem: to study the environmental conditions of the facilities where the application will run, the light technology and the wattage of the chosen lamps, nowadays we are moving to LED technology due to the increased life and absence of flicker, among other benefits. Nevertheless, it would be desirable to make the intelligent system more robust to lighting conditions changes, as in the case of texture classification systems [1]. On such systems the effect of light changes on the measured features may eventually lead to texture misclassification and performance degradation. In this paper we present an approach that will be helpful to overcome such problems when the light comes from a directional source, such as halogen projectors, LED arrays, etc.

This paper presents a scheme for image decomposition and reconstruction, using complex Gabor wavelets. Gabor functions have been used extensively in areas related to the human visual system due to their localization in space and bandlimited properties. However, since the standard two-sided Gabor functions are not orthogonal and lead to nearly singular Gabor matrices, they have been used in the decomposition and feature extraction of images rather than in image reconstruction. In an attempt to reduce the singularity of the Gabor matrix and produce reliable image reconstruction, in this paper, we show that a single-sided Gabor function can accomplish both, with the reconstruction residual error being very small (PSNR of at least 300 dB).