Catálogo de publicaciones - libros

DTV data broadcasting offers broadcasters the new business model to acquire considerable revenue from digital broadcast. When the broadcasters integrate and maintain the data broadcast station system, they encounter the cost to be paid unnecessarily because there is no common interface specification among the equipments of DTV station system. This paper describes the flexible and open interface that will decrease the wasteful expenses in moving the transition to data broadcast station system.

We previously presented a demosaicking method that simultaneously removes image blurs caused by an optical low-pass filter used in a digital color camera with the Bayer’s RGB color filter array. Our prototypal sharpening-demosaicking method restored only spatial frequency components lower than the Nyquist frequency corresponding to the mosaicking pattern, but it often produced ringing artifacts near color edges. To overcome this difficulty, this paper introduces the super-resolution into the prototypal method. First, we for mulate the recovery problem in the DFT domain, and then introduce the super-resolution by the total-variation (TV) image regularization into the sharpening-demosaicking approach. The TV-based super-resolution effectively demosaics sharp color images while preserving such image structures as intensity values are almost constant along edges, without producing ringing artifacts, but it tends to flatten signal variations excessively in texture image regions. To remedy the drawback, furthermore we introduce a spatially adaptive technique that controls the TV image regularization according to the saliency of color edges around a pixel.

Image completion is a method to fill the missing portions of an image caused by the removal of one or more foreground or background elements. In this paper a novel image completion algorithm is proposed for removing significant objects from natural images or photographs. The completion is realized in the following three steps. First, a gradient-based model is presented to determine the gradient-patch filling order. This step is critical because a better filling order can improve the continuation of image structures. Second, we implement the gradient-patch update strategy by measuring the exponential distance between the source patch and the target one in gradient domain. In order to find a better patch matching and propagating algorithm, we incorporate the gradient and color information together to determine the target patch. Third, a complete image is achieved by solving the Poisson equation with the updated image gradient map. Some experimental results on real-scene photographs are given to demonstrate both the efficiency and image equality of our novel method.

Motion estimation is the biggest part of any video encoding system since it could significantly affect the encoding time and output quality. Until now, several motion estimation algorithms that have the fast motion estimation and a good PSNR are proposed. However these are not considered an embedded system processor using SDRAM as a video frame memory. In this paper, we propose new fast predictive block matching algorithm, named predictive directional rectangular zonal search . Comparing with existing algorithms, it has not only similar output quality, but more efficient operating with SDRAM. It makes possible to design a hardware motion estimator into a compact one. A motion estimation hardware block that is based on the proposed algorithm for H.263 and MPEG-4 video encoder, is implemented by 0.3 μm 1.2 V CMOS technology.

In this paper, we propose a method to refine a motion field from image sequences and region-based motion segmentation using the motion information. An initial motion field is generated by a block matching algorithm. We compute the motion profile at each block and define the motion confidence measure from the motion profile. In the refining process, we regulate the motion vectors with low confidence to those with high confidence. In the segmentation stage, each frame of the image sequence is partitioned into regions by a watershed algorithm and a motion vector is assigned to each region. After constructing a region adjacency graph, the graph is segmented by the normalized cuts algorithm. The experiments show that the proposed method provides satisfactory results in motion segmentation from image sequences with or without camera motion.

A motion adaptive de-interlacing technique with horizontal and vertical motions detection is proposed and its performances are examined. Object movement happens quite often in film broadcasting and normally they move horizontally, vertically, or diagonally. The movements tend to destabilize the quality of performance such as jagged effect, blurred effect, and artifacts effect, while de-interlacing technique is utilized. In our proposed method, de-interlacing begins with object motion detection, which is to ensure that the interfield information is used precisely. The proposed method also utilizes intrafield de-interlacing by median edge dependent interpolation, Median EDI, while the object movement is not detected. The simulation results show that the proposed algorithm exhibits better performances than other interpolation algorithms.

This paper describes a method of image generation based on transformation integrating multiple differently focused images. First, we assume that objects are defocused by a geometrical blurring model. And we combine acquired images on certain imaging planes and spatial frequencies of objects by using a convolution of a three-dimensional blur. Then, we reconstruct an all-in-focus image from the acquired images based on spatial frequency analysis using three-dimensional FFT. Some experiments of image generation utilizing synthesized images and real images are shown and extension of the method integrating multiple differently focused images in three-dimensional frequency domain is discussed.

This research is to make a series of NURBS surfaces for the virtual 3D conceptual design and the styling process by applying arbitrary free-hand strokes. The surface can be modified in real-time calligraphic stroke based free form deformation. The suggested algorithm is used to create 3D NURBS surfaces for styling object using free-hand strokes with the posture information of the input wand. The algorithm presented in this paper can help product designers in the conceptual engineering stage, even if he or she has no idea about the shape of a target product.

This paper presents a new method of line drawing based on the hypothesis that artists draw the lines that decompose the object into parts, and the lines that help convey the shapes of the parts. But they draw these lines differently depending on the viewpoint. Contours are the most obvious part-decomposing lines. Valley lines, which typically delimit convex parts, are also part-decomposing lines. As shape-conveying lines, ridge lines on each part are chosen; they are good at conveying the shape of parts in that they are maxima of the principal curvatures on the part surface. So, valley and ridge lines are good candidates in line-drawing. But they have been dismissed because they are view-independent unlike contours. But because of their shape-conveying capability, they have a strong intuitive appeal as candidates for line-drawing. So we propose a way to ”redeem” them by making them view-dependent: Valley and ridge lines are given strengths depending on how the view direction relates to the surface normals to the lines. On the other hand, when valleys and ridges are extremely strong, for example, when they are sharp edge lines, they are drawn regardless of viewpoint. We have found that the view-dependent valley and ridge lines are quite stable with respect to viewpoint change.

The paper presents an efficient way of designing lighting setup for rendering 3D face model. Specifically, we focus on obtaining lighting direction for Rembrandt lighting. A Rembrandt patch is a triangle defined as the bright region surrounded by self and cast shadows on a check area, and we use the self- and cast-shadow curves for computing the direction of main lighting. A user graphically specifies a Rembrandt patch on a 3D model. From the user input, lighting directions are estimated from the cast- and self-shadow geometry on a 3D face model. The final lighting direction is decided among the candidates predicted by the self and cast shadows. The presented method lets a user interactively design and achieve Rembrandt lighting by alleviating repetitive manual search for the light direction by trial and error. Experimental results show the effectiveness of the presented method. It suggests appropriate Rembrandt lighting directions quickly and easily.