Corresponding geometric distorted images using an uncertainty inference method

We present further results of the application of Dempster-Shafer theory for uncertainty reasoning in corresponding distorted images in computer vision. In a previous work (J.D.S. Silva et al., 2002), the model was applied to correspond radiometrically distorted images, that is, images with differences in brightness and contrast, as an extension of the work developed by J.D.S. Silva et al. (2001). The results showed the model is robust when dealing with pairs of nonequalized images and encouraged us to try to correspond geometric distorted images, that is, pairs of images in which one is rotated in relation to the other. In the conducted experiments, the right image was rotated by different angles to simulate the desired geometric distortions. The model was applied to a pair of rotated images and it successfully established the correspondence of a pair of points. As in previous works, the correspondence evidences are based on the contextual and structural features of the points, and their combination is performed by Dempster-Shafer's rule of combination for uncertainty reasoning. A search process maximizes the belief on the combined evidences.