Texture feature extraction and indexing by Hermite filters

We present a texture feature extraction for image indexing and retrieval based on Gabor-like Hermite filters. These ones satisfy a frequency constraint of steered discrete Hermite filters, which form a local orthogonal basis and agree with the Gaussian derivative model of the human visual system. Fast implementation of such filters is performed by a normalized recurrence relation of their discrete representation, the Krawtchouk filters. In order to achieve dimensionality reduction for texture image indexing purposes, we apply a compact parametric texture model, which corresponds to the spatial autocorrelation of each subband output. Experimental results obtained from a texture image database are also presented.