Fast filter bank synthesis by periodical anamorphosis

We present a method for analysing spatial frequency and orientation, based on a space variant (SV) filter bank. Frequency response of each filter depend on its spatial position. Theses filters are inspired by retino-cortical map and describe a model ("Pinwheel") of orientation column of striate cortex. These SV filters result from the combination of two transformations: a periodical projection and a space invariant filtering. This special case of implementation of SV filtering brings a computed time gain in comparison to classical filter bank such as the wavelets Morlet 2D. After description of the method, we explain the different parameters which optimise the frequency cover. It appears that the filter bank constitutes good attributes for texture segmentation.