Contributions à la caractérisation & structuration d’images et de séquences vidéo
Youssef Chahir · HAL (Le Centre pour la Communication Scientifique Directe) · 2012
The paper presents a summary of work in sequence analysis of images and video.In the first part, I present my work in spatio-temporal regularization based graphs to resolve open issues in image processing: the restoration of noisy images and sequences of missing data, and the classification of data in images. Several applications are proposed.To do this, we use some properties of the graph (energy, ...) as well as its intrinsic space obtained by a nonlinear dimensionality reduction analysis of random walks through the space.The second part is devoted to the analysis and characterization of video sequences. At first, I present a hybrid model of skin based on induction graphs and my work related to the convex segmentation by volume of a moving object using active contours, and optical flow. Then, a section is devoted to work from multidisciplinary projects related to the analysis of gestures and gaze. Databases share specific videos are analyzed using discrete regularization on graphs and spatio-temporal moments such as Zernike ....The last theme is dedicated to the analysis of real-time look where we focus on the prediction of gaze using Gaussian processes.