Rotation-invariant texture analysis using multi-resolution Slit kernels

Abstract

This paper presents a set of simple Slit kernel suitable for regular and rotation-invariant texture analysis. The fast algorithm used to compute its feature is also presented. Extracting Slit features requires 11 to 17% of arithmetic operations when compared to that for Gabor features. Experimental results for the classification of rotated texture images indicate that Slit kernels perform as well as, or sometimes better than, Gabor ones. This makes Slit suitable for applications where computational speed is important.