| Object recognition |
Object recognition showcaseLeader: Nicu Sebe, UVA
Goals: The objective of the showcase is to create a demo of object recognition including object localization in the image. The demo will be trained on about 10-20 objects. When an image is loaded into the demo system, it will first attempt to recognize which of the objects on which it has been trained is in the image (using object recognition technology from UVA & INRIA) and then attempt to mark the region of the image which contains the object. We are going to use the novel approach for retrieval of object categories developed by INRIA-IMEDIA and UVA, based on a novel type of image representation: the Generalized Correlogram (GC). In this image representation, the object is described as a constellation of GCs where each one encodes information about some local part and the spatial relations from this part to others (i.e., the part's context). We will integrate the representation by boosting the system in order to obtain a compact model that is represented by very few features, where each feature conveys key properties about the object's parts and their spatial arrangement. We aim at achieving processing times of less than 10 seconds for an image. Current Status: We have performed an investigation on different salient point detection algorithms. Observing that most current methods use only the luminance information of the images, we investigated the use of colour information in interest point detection. To determine the characteristic scale of an interest point, we developed a new colour scale selection method. The preliminary results show that using colour information and boosting salient colours yields improved performance in retrieval and recognition tasks. These results were reported in a presentation at the Computer Vision Winter Workshop and in a submission to the International Conference on Image Retrieval. Currently, we are preparing a publication for International Conference on Computer Vision. |