Unsupervised object extraction by contour delineation and texture-based discrimination

An unsupervised object extraction system capable of separating animal images from background in natural scenes has been developed. Edges are detected from original images at multiple reduced resolutions and used to identify object contours. In order to determine the inside and the outside of an object, texture information is compactly represented using oriented edges and analyzed. The contour information and the texture information are integrated so that they complement each other. As a result, the object region has been successfully extracted as is from the scene with a fairly-well defined boundary line. Simulation experiments were carried out on several natural images and promising performance has been demonstrated.