Catálogo de publicaciones - libros

In this paper, an automatic technique for natural image matting is proposed. We use visual characteristics of the background and foreground regions for matting. Through the Locally Linear Embedding (LLE) of high-dimensional features, we estimate foreground and background color of unknown pixels. We use gradient information and a hierarchical model to enhance the proposed matting technique. Instead of a user interaction for tri-map generation, we propose an automatic technique that obtains a tri-map using a multi-view camera. A reliability map for the depth of a scene facilitates the generation of a tri-map. It is proven that feature sets obtained from training images can be applied to similar images or video frames.

For computation of the epipolar geometry from central- omni-directional images, the use of the spherical camera model is essential. This is because the central-omnidirectional cameras are universally expressed as the spherical camera model when the intrinsic parameters of the cameras are calibrated. Geometrically, for corresponding points between two spherical images, there exists the same epipolar constraint as the conventional pinhole-camera model. Therefore, it is possible to use the conventional eight-point algorithm for recovering camera motion and 3D objects from two spherical images. In this paper, using the geometric properties on rotation of the spheres, we propose a method of the accurate computation based on the rectification of the spherical-camera images via the conventional eight-point algorithm.

The following paper presents a method that allows for a parallel implementation of the most computationally expensive element of the deformable template paradigm, which is a grid-matching procedure. Cellular Neural Network Universal Machine has been selected as a framework for the task realization. A basic idea of deformable grid matching is to guide node location updates in a way that minimizes dissimilarity between an image and grid-recorded information, and that ensures minimum grid deformations. The proposed method provides a parallel implementation of this general concept and includes a novel approach to grid’s elasticity modeling. The method has been experimentally verified using two different analog hardware environments, yielding high execution speeds and satisfactory processing accuracy.

This paper presents a simple method for segmenting text image on the basis of color components. It is shown how segmentation can benefit from splitting color signals into chromatic and achromatic components and separately smoothing them by proposed clustering method. We analyze and compare the performance of several color components in terms of segmentation of the text regions from color natural scenes. We also perform a fast 1-dimensional -means clustering algorithm. Therefore we can perform accurate object segmentation using both H and I components. And then, the effectiveness and reliability of proposed method are demonstrated through various natural scene images. The experimental results have proven that the proposed method is effective.

This paper presents a model-based approach to monocular tracking of human body using a non-calibrated camera. The tracking in monocular images is realized using a particle filter and an articulated 3D model with a cylinder-based representation of the body. In modeling the visual appearance of the person we employ appearance-adaptive models. The predominant orientation of the gradient combined with ridge cues provides strong orientation responses in the observation model of the particle filter. The phase that is measured using the Gabor filter contributes towards strong localization of the body limbs. The potential of our approach is demonstrated by tracking of the human body on real videos.

In this paper, we propose a method that measures the height and stagger of an overhead electric-railway line using the stereo images. Two 1624 X 1236 pixel area scanner CCD cameras are used. To quickly and accurately extract, from a photographed image, the area of the overhead line on which the line laser is shone, we consider the established fact that the overhead line is the lowest among the electric wires. And to precisely measure the height and stagger in low resolution, sub-pixel and line fitting methods are used. Also, because of the different pixel resolution of the camera according to the overhead line position, we compensate the measurement result through camera calibration. We aimed for a measurement accuracy of 1mm error and indeed our experimental results show that the proposed method achieves that.

This study uses object-extracting technique to extract thumb, index, middle, ring, and small fingers. The algorithm developed in this study can locate the precise locations of fingertips and finger-to-finger-valleys. The extracted fingers contain many useful geometry features. One can use these features to identify fingers. The geometry descriptor is used to transfer geometry features of fingers to another feature-domains. Fingers are scaled to allow fingers possess more salient features. Finger is also magnified by the basis of “distance multiplying gray level”. After finger magnifying, finger will possess more salient feature. Image subtraction is used to examine the difference of the two fingers.

This paper presents a method to suppress the bias artifact, also known as RF-inhomogeneity, in Magnetic Resonance Imaging (MRI). This artifact produces illumination variations due to magnetic field fluctuations of the device. In the latest years many works have been devoted to face this problem. In this work we present the 3D version of a new approach to bias correction, which is called Exponential Entropy Driven Homomorphic Unsharp Masking ( − ). This technique has been already presented by some of the authors for the 2D case only. The description of the whole method is detailed, and some experimental results are reported.

Color-based particle filter for object tracking has been an active research topic in recent years. Despite great efforts of many researchers, there still remains to be solved the problem of contradiction between efficiency and robustness. The paper makes an attempt to partially solve this problem. Firstly, the is introduced by which histogram of any rectangle region can be computed at negligible cost. However, straightforward application of the Integral Histogram Images causes the problem of “curse of dimensionality”. In addition, traditional histogram is inefficient and inaccurate. Thus we propose to adaptively determine histogram bins based on K-Means clustering, which can represent color distribution of object more compactly and accurately with as a small number of bins. Thanks to the Integral Histogram Images and the clustering based color histogram, we finally achieve a fast and robust particle filter algorithm for object tracking. Experiments show that the performance of the algorithm is encouraging.

We describe an algorithm for fitting a Catmull-Clark subdivision surface model to an unstructured, incomplete and noisy data set. We complete the large missing data regions with the shape information and produce a smooth, compact and structured data description. The result can be used for further data manipulation, compression, or visualisation. Our fitting algorithm uses a technique which manipulates the base mesh of the subdivision model to achieve better approximation. We extend the approach designed for scientific visualisation and animation to deal with incomplete and noisy data and preserve prior shape constraints where data is missing. We illustrate the algorithm on range and stereo data with a set of different subdivision models and demonstrate the applicability of the method to the problem of novel view synthesis from incomplete stereo data.