Detecting Perceptually Salient Texture Regions in Images

摘要

Texture or patterns on objects are often used as cues for quickly selecting important aspects of a scene. Such a selection can be important for a variety of tasks including object recognition, tracking, surveillance, and image indexing of databases. In this paper, we examine the problem of perceptual grouping based on texture saliency and demonstrate its use in selecting salient texture regions in natural scenes. Specifically, we propose that texture saliency can be inferred from the relative contrast among the bright and dark regions in the texture. In particular, a texture saliency measure is derived that extracts relatively bright and dark regions within a texture and analyzes their interplay in terms of perceptually salient attributes. A texture region segmentation is also presented to enable the application of texture saliency to texture regions in scenes. The performance of texture-based region selection in the ranking of regions is demonstrated.