A novel texture clustering method based on shift invariant DWT and locality preserving projection

Abstract

We propose a novel texture clustering method. A classical type of (approximate) shift invariant discrete wavelet transform (DWT), dual tree DWT, is used to decompose texture images. Multiple signatures are generated from the obtained high-frequency bands. A locality preserving approach is applied subsequently to project data from high-dimensional space to low-dimensional space. Shift invariant DWT can represent image texture information efficiently in combination with a histogram signature, and the local geometrical structure of the dataset is preserved well during clustering. Experimental results show that the proposed method remarkably outperforms traditional ones.