Region Growing in Textured Outdoor Scenes.

Abstract : Some progress has been made in scene analysis in finding boundaries between textured objects. However, standard line-finding and region growing algorithms have not been able to deal with strong local variations of intensity and color that repeat themselves more globally in images of natural scenes. These problems are significant with respect to both micro-texture, say the leaves of a tree, as well as macro-texture, the light green of leafy branches vs. dark green or black shadows between branches in a tree. How can local variations be globally bound together into a cohesive region for perceptual identification. The approach that is being examined involves the reduction in size of the image while extracting features such as intensity, color, and color variation in local windows across the image.