Building Efficient Framework on Visual Object Recognition

The problem of recognizing generic object categories is one of most active areas in computer vision and pattern recognition. The problem lies in the center of visual object recognition challenge. The task is trying to automatically recognition genetic object in natural scene. The problem statement: It views whole image as one item, aim at extracting the most impressive Region of Interests (ROI) to be classified. It is challenging due to: large data set of various numbers of classes, even under the same class, the shape and appearance may vary at different levels, in addition, real world images naturally contain irrelevant background, and object occlusions and large viewpoint changes may also result in the obstacle. How to handle this problem efficiently and obtain promising recognition rate is still in progress, and the improvement on this work will make contribution to the fundamental computer vision applications.