Learning simple texture discrimination filters
Everyday tasks like walking on the street, recognizing a friend or understanding a scene seem so simple and immediate that transposing it to a computer might seem like an easy task. Only when we try it do we realize our immense talent, as humans, in making sense of the data that reaches our senses. In this talk I illustrate some of these difficulties and particularize for the context of texture discrimination. I introduce a simple supervised learning approach (using Genetic Algorithms) that enables high-frame rate texture discrimination and compare it with current state-of-the-art methods. I further particularize the general methodology to rotationally discriminant and rotationally invariant discrimination. I conclude with experimental results, which illustrate that it is successful in capturing the essence of the texture discrimination problem.