Illusions sometimes give us the keys to understand the mechanisms of the visual information processing. But, the mechanisms of the brightness illusions, for example the Mach band of the Craik-O'Brien effect, are not clear yet. In this paper, the role of the filling-in process on the brightness perception is explained. Its computational theory is discussed. And a neural network model of the filling-in process on the brightness perception is constructed based on the computational theory. Some properties of the Mach band and the Craik-O'Brien effect are simulated with the model, and its behavior is similar to some psychological experiment results.
Automatic face recognition is one of the most important subjects in human image processing, aiming toward intelligent human computer interaction. This paper presents an outline of an approach to achieving a face recognition system, along with some early experimental results. Assuming that face size and position in the image will be standardized using the results of image segmentation, ordinary 2D pattern matching is used to classify face images. Pattern matching parameters giving the best classification results are also discussed.
Instead of frequent appearance of partial occlusion among objects in a scene, inferring and recovering the shapes of occluded parts have been left untouched as one of the difficult issues in computer vision. We describe an approach to the issue through the determination of partial ordering of occlusion of objects from T-type junctions on the contours of regions. We propose a scheme of inferring the shape of occluded parts based on the assumption of smoothness of edges of objects.
Basic concepts defined for understanding of optical illusions include a position-dependent PSF (point spread function) and an assumed measure of fixation over the pattern plane. An exact simulation of the famous Hering's illusion has been carried out by employing the position-dependent PSF. Reasonable understanding of the Mueller-Lyer's and Poggendorff's illusions has been achieved in terms of the assumed impressibility measure of fixation. The well-known laterally inhibitory function has been included in the mathematical description of the position-dependent PSF.