Catálogo de publicaciones - libros

Vector median filtering is a well known technique for reducing noise in color images. These filters are defined on the basis of a suitable distance or similarity measure, being the most common used the Euclidean and City-Block distances. In this paper, a Fuzzy Metric, in the sense of George and Veeramani (1994), is defined and applied to color image filtering by means of a new Vector Median Filter. It is shown that the standard Vector Median Filter is outperformed when using this Fuzzy Metric instead of the Euclidean and City-Block distances.

In this paper, we present new wavelet shrinkage methods for image denoising. The methods take advantage of the higher order statistical coupling between neighboring wavelet coefficients and their corresponding coefficients in the parent level. We also investigate a multiplying factor for the universal threshold in order to obtain better denoising results. An empirical study of this factor shows that its optimum value is approximately the same for different kinds and sizes of images. Experimental results show that our methods give comparatively higher peak signal to noise ratio (PSNR), require less computation time and produce less visual artifacts compared to other methods.

This paper addresses a novel video summarization procedure that produces a dynamic (video) abstract of the original video sequence. To remain temporal characteristic in the video abstract, a newly time-oriented feature is introduced. The approach relies on time instances, frame rate of the original video sequence and display speed of the video summary to select frames for a video abstract. This method preserves the time-evolving dynamic nature of the video content. Based on a input video sequence, real-time video summarization can be achieved at different display speed.

Shadow removal is of significance in computer vision. Shadow usually adheres to texture, and it is difficult to decouple shadow from texture without any assumptions. In this paper, we suppose that the texture beneath shadow should bear the similar patterns, which makes the problem of simultaneously recovering the reflectance and light attenuation factor solvable with sole image. A novel shadow removal method is proposed, which is fulfilled by complementing the lost radiance in shadow region. Experiments show that our proposed approach could maintain the texture structure and the coherent lighting with the intact region.

The weighted sum of squared differences cost function is often minimized to align two images with overlapping fields of view. If one image is shifted with respect to the other, the cost function can be written as a sum involving convolutions. This paper demonstrates that performing these convolutions in the frequency domain saves a significant amount of processing time when searching for a global optimum. In addition, the method is invariant under linear intensity mappings. Applications include medical imaging, remote sensing, fractal coding, and image photomosaics.

In this paper, we propose a blur identification for fully digital auto-focusing (FDAF) techniques under the assumption of isotropic point spread function (PSF). The proposed blur identification starts with the estimation of one-dimensional (1D) step response, and estimate the two-dimensional (2D) PSF using the fundamental relationship between the step and the impulse responses. For more accurate blur identification, we present the PSF interpolation method that fills out the PSF element off the concentric circle. We also propose least squares 2D PSF solutions that robustly estimates the strength of the concentric circles under noisy, incomplete observation environment. Experimental results prove that the proposed blur identification method outperforms the existing ones in the sense of both accuracy and efficiency.

In this paper, the Mumford-Shah (MS) model and its variations are studied for image segmentation. It is found that using the piecewise constant approximation, we cannot detect edges with low contrast. Therefore other terms, such as gradient and Laplacian, are included in the models. To simplify the problem, the gradient of the original image is used in the Rudin-Osher-Fatemi (ROF) like model. It is found that this approximation is better than the piecewise constant approximation for some images since it can detect the low contrast edges of objects. Linear approximation is also used for both MS and ROF like models. It is found that the linear approximation results are comparable with the results of the models using gradient and Laplacian terms.

A robust digital image stabilization algorithm is proposed using a Kalman filter-based global motion prediction and phase correlation-based motion correction. Global motion is basically estimated by adaptively averaging multiple local motions obtained by phase correlation. The distribution of phase correlation determines a local motion vector, and the global motion is obtained by suitably averaging multiple local motions. By accumulating the global motion at each frame, we can obtain the optimal motion vector that can stabilize the corresponding frame. The proposed algorithm is robust to camera vibration or unwanted movement regardless of object’s movement. Experimental results show that the proposed digital image stabilization algorithm can efficiently remove camera jitter and provide continuously stabilized video.

In this paper, a localized wavelet thresholding strategy which adopts context-based thresholding operators is proposed. Traditional wavelet thresholding methods, such as VisuShrink, LevelShrink and BayesShrink, apply the conventional hard and soft thresholding operators and only differ in the selection of the threshold. The conventional soft and hard thresholding operators are point operators in the sense that only the value of the processed wavelet coefficient is taken into consideration before thresholding it. In this work, it will be shown that the performance of some of the standard wavelet thresholding methods can be improved by applying a localized, context-based, thresholding strategy instead of the conventional thresholding operators.

In the classical image processing paradigm, the fundamental idea of image enhancement is to produce a new image such that it exposes information for analysis more than the original image. Fuzzy Logic methods are one of valuable and frequently used techniques among many other image enhancement approaches. In this work we demonstrate that a type-2 fuzzy logic system is able to perform image contrast enhancement better than its type-1 counterpart. Examples are provided. Results are discussed.