A psychophysical study of dominant texture detection

Images of everyday scenes are frequently used as input for texturing 3D models in computer graphics. Such images include both the texture desired and other extraneous information. In our previous work [Lu et al. 2009], we defined dominant texture as a large homogeneous region in an input sample image and proposed an automatic method to detect dominant textures based on diffusion distance manifolds. In this work, we explore the identification of cases where diffusion distance manifolds fail, and consider the best alternative method for such cases.