Catálogo de publicaciones - libros

We address the problem of super resolved generation of novel views of a 3D scene with the reference images obtained from cameras in general positions; a problem which has not been tackled before in the context of super resolution and is also of importance to the field of image based rendering. We formulate the problem as one of estimation of the color at each pixel in the high resolution novel view without explicit and accurate depth recovery.We employ a reconstruction based approach using MRF-MAP formalism and solve using graph cut optimization. We also give an effective method to handle occlusion. We present compelling results on real images.

The paper proposes a novel approach for classification of sports images based on the geometric information encoded in the image of a sport’s field. The proposed approach uses invariant nature of a cross-ratio under projective transformation to develop a robust classifier. For a given image, cross-ratios are computed for the points obtained from the intersection of lines detected using Hough transform. These cross-ratios are represented by a histogram which forms a feature vector for the image. An SVM classifier trained on aprior model histograms of cross-ratios for sports fields is used to decide the most likely sport’s field in the image. Experimental validation shows robust classification using the proposed approach for images of Tennis, Football, Badminton, Basketball taken from dissimilar view points.

The paper is focused on the problem of pose estimation from a single view in minimum conditions that can be obtained from images. Under the assumption of known intrinsic parameters, we propose and prove that the pose of the camera can be recovered uniquely in three situations: (a) the image of one circle with discriminable center; (b) the image of one circle with preassigned world frame; (c) the image of any two pairs of parallel lines. Compared with previous techniques, the proposed method does not need any 3D measurement of the circle or lines, thus the required conditions are easily satisfied in many scenarios. Extensive experiments are carried out to validate the proposed method.

In this paper, we propose a new face recognition approach based on local binary patterns (LBP). The proposed approach has the following novel contributions. (i) As compared with the conventional LBP, anisotropic structures of the facial images can be captured effectively by the proposed approach using elongated neighborhood distribution, which is called the elongated LBP (ELBP). (ii) A new feature, called Average Maximum Distance Gradient Magnitude (AMDGM), is proposed. AMDGM embeds the gray level difference information between the reference pixel and neighboring pixels in each ELBP pattern. (iii) It is found that the ELBP and AMDGM features are well complement with each other. The proposed method is evaluated by performing facial expression recognition experiments on two databases: ORL and FERET. The proposed method is compared with two widely used face recognition approaches. Furthermore, to test the robustness of the proposed method under the condition that the resolution level of the input images is low, we also conduct additional face recognition experiments on the two databases by reducing the resolution of the input facial images. The experimental results show that the proposed method gives the highest recognition accuracy in both normal environment and low image resolution conditions.