Non causal adaptive quadratic filters for image filtering and contrast enhancement

In image contrast enhancement [2, 4, 5], quadratic and more generally polynomial filters are a very popular class of nonlinear filters. These filters exhibit good performances in terms of visual quality, but present some drawbacks such as the elimination of usefull information when using a fixed filter. In this paper we propose a new family of adaptive quadratic filters, where a weighted filter mask is adaptively determined according to the minimization of a prediction error. This filter is then used to enhance locally the image contrast. The results we proposed point out the improvement provided by these new filters in comparison with recent approaches [4, 2].