DESARROLLO DE UNA APLICACIÓN DE RECUPERACIÓN DE OBJETOS Y ANÁLISIS DE LOS DESCRIPTORES LOCALES INVARIANTES UTILIZADOS

Abstract

Retrieval of objects in big datasets of images and videos is one of the tasks of the European Project ASASEC (Advisory System Against Sexual Exploitation of Children). Spanish National Police Department of Technological Information is interested in selecting a box around an object considered as evidence and finding the same object in a great collection of images and videos registered as evidences of crimes of sexual exploitation of children. For this purpose, an application that allows the user to choose one image, select a box over the image using the mouse and look for similarities between the region of interest and the rest of the images in a folder has been developed. In order to describe the region of interest and the rest of the images, several well-known robust local invariant descriptors have been used such as: SURF, SIFT and COSFIRE. Experiments were performed on a 100 images dataset taken from videos recorded in the same room, Lonelygirl dataset, with different poses of the camera, occlusions and illumination changes. Results show the benefit of using SIFT and COSFIRE opposite to SURF and HOG.