Computer Vision Talks at Columbia University

Perceptually-Driven Statistical Texture Modeling

Eero Simoncelli

Center for Neural Science and Courant Institute for Mathematical Sciences, NYU

Thursday, April/18, 2:30 PM

Interschool Lab, 7th Floor CEPSR, Schapiro 

Host: Prof. Shree Nayar 

Abstract 

We've recently developed a parametric statistical model of visual texture that is based both on empirical study of natural texture images, and the processing that occurs in early stages of the human visual system (Portilla & Simoncelli, IJCV, 30(1):49-71, 2000). The model is parameterized by a set of statistics computed on pairs of coefficients corresponding to basis functions at adjacent spatial locations, orientations, and scales. We develop an efficient algorithm for synthesizing random images subject to these constraints, by iteratively projecting onto the set of images satisfying each constraint, and use this to test the perceptual validity of the model. In particular, we demonstrate the necessity of subgroups of the parameter set by showing examples of texture synthesis that fail when those parameters are removed from the set. We also demonstrate the power of our model by successfully synthesizing examples drawn from a diverse collection of artificial and natural textures. Finally, we show applications of this model to image extrapolation and enhancement.