Purchase this article with an account.
Lei Liu, Janis M. White; A computational model for discrimination of even and random textures. Journal of Vision 2002;2(10):133. doi: https://doi.org/10.1167/2.10.133.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Julesz, Gilbert and Victor (1978) developed the even and random textures to demonstrate that pre-attentive texture discrimination did not have to rely on differences in global Fourier characteristics. We developed a computational model that was based on local features in the texture. This model consists of four steps: 1) extracting rectangular features (textons) of different sizes and aspect ratios from two input texture patches; 2) calculating a density difference for each type of texton; 3) pooling density differences of all types of textons to form a decision variable; 4) determining whether the two input patches belong to the same or different texture using a threshold mechanism. The behavior of the model was compared with human performance in the discrimination of two types of degraded textures. In the first type of degradation, randomly chosen checks of even and random textures were deleted. Normal observers could maintain near perfect discrimination when the proportion of deleted checks was less than 40%. Further deletion resulted in a rapid deterioration of discrimination performance. Even and random textures became indiscriminable when 60% of checks were deleted. In the second type of degradation, a portion of the checks were deleted first, and then the positions of the remaining checks were scrambled so that they could take the places of deleted checks. Position scrambling appeared to have a great effect on texture discrimination. Textures became indiscriminable when 10% of checks were deleted and the positions of the remaining checks were scrambled. The behavior of the model was obtained by giving pairs of equally degraded texture patches to the model as input, and recording the model's decisions. The model exhibited behavior similar to human observers in the discrimination of degraded even and random textures.
Julesz, B., Gilbert, E.N., & Victor, J.D. (1978). Visual discrimination of textures with identical third-order statistics. Biological Cybernetics, 31, 137–140.
This PDF is available to Subscribers Only