With all the technological advances in the field of computer vision, one of the greatest challenges has been getting a computer to recognize an image. Various methods exist for shape based object recognition when a whole object contour is given. This project addresses the next step in computer vision: shape-based object recognition based on contour parts.
Contour part of a horse.
Given a relatively small object part, humans can recognize an object when the part is sufficiently unique. For example, it is obvious to us that the figure at right shows a horse. Moreover, the ease with which humans recognize articulated objects suggests that shape recognition in humans may solely be based on parts.
Past projects in the area of computer vision have mainly focused on using entire contours to recognize an image. Because of the need for an entire contour, these methods are incapable of handling the real-world problems of recognizing objects in images: segmentation errors, overlap, and occlusion. The few past projects that have focused on recognition by parts have struggled with great obstacles such as noise, distortion, scale variation, and segmentation errors. In this project, we take the parts-recongition trait of human vision and replicate it for use in computer vision systems. By isolating particulary promising contour parts of a digital image and processing those parts against a database of known contour parts, we find that we can classify an object in a digital image with a high degree of accuracy.
The goal of this project is to recognize and classify contour parts extracted from images. We use partial shape contours to identify the presented shape with shapes of similar natures and classifications. This is done efficiently and effectively using such tools as Discrete Curve Evolution and Shape Contexts. Shapes can even be isolated and identified despite noise, distortion, scale differential, and occlusion.
| Motivation | Next -> |