Seeing overall structures in preattentive vision

Abstract

According to the recent psychological researches by Julesz[1-3], Beck[4-6], and Treisman[7], human visual system operates in two distinct modes: preattentive mode and attentive mode. Preattentive vision detects local conspicuous features, called textons and directs attentive vision to the location where texton-gradients occur. The preattentive texture discrimination is the result of difference in first-order statistics of textons. Based on this theory, several computational models have been built to compute texture boundaries by computing local features and their local distributions [8-11]. Evidences presented in this paper show that preattentive texture discriminability is not merely determined by the first-order statistics of textons, but it is also under the influence of the overall structures of the textures. The overall structures of the textures which can be preattentively perceived are built on the local lightness or colors rather than the local spatial features or textons. Therefore preattentive vision is not merely engaged in detecting local features, it can also grasps the overall structures based on the variations of local lightness or colors.