| Based on the oscillatory correlation approach to the image segmentation problem, we study texture and motion segmentation. In oscillatory correlation, each segmented region is represented by an synchronized oscillator population and different segmented regions are represented by oscillator populations that desynchronize from one another. In other words, synchronization and desynchronization implicitly perform region labeling. Our method to each case study is composed of two parts. The first part extracts local features, e.g. texture, motion, luminance, from input scenes. The second part is a LEGION (Locally Excitatory Globally Inhibitory Oscillator Networks) architecture, which computes oscillatory correlation.; In the texture-based method, the first part determines a novel set of texture features derived from a Gaussian Markov Random Fields (GMRF) model. Unlike a general GMRF-based approach, our method does not employ model parameters as features or require the extraction of features for a fixed set of texture types selected a priori.; In the motion-based method, segmentation is performed by integrating motion and luminance information. The method is composed of two parallel pathways that process motion and luminance, respectively. Inspired by the visual system, the motion pathway has two stages. The first stage estimates local motion at locations with reliable information. The second stage performs segmentation based on local motion estimates. In the luminance pathway, the input scene is segmented into regions based on its luminance distribution. Subsequently, segmentation results from the two pathways are integrated to refine motion estimates. The integration stage is composed of three steps including an occlusion analysis, the elimination of unreliable estimates, and motion filling-in within homogeneous regions. The final segmentation is performed based on the refined estimates in the motion pathway. Unlike the networks in the textured-based method and the luminance pathway, a multilayer LEGION network is employed in the motion pathway to facilitate the representation of multiple motions in the same area.; We demonstrate the performance of the two methods using both synthetic and real images. For the motion-based method, we also consider visual stimuli used in psychophysical experiments. The locality in feature extraction and the biological plausibility of our computational approach, namely oscillatory correlation, suggest new future directions in the understanding of the human visual system. |
