Colour local pattern: a texture feature for colour images

Abstract

The Julesz’s conjectures were the foundations for the development of many methods for texture discrimination as spatial arrangements of local patterns. These local patterns represent a bridge between a first conjecture that handles a purely statistical approach and the texton theory. Beyond the mathematical construction of the Local Binary Patterns (LBP) proposed by Ojala, the proposed pattern defined by a circular neighbourhood can be considered as texton. Nevertheless, the binarisation step inside the Local Binary Pattern reduced the ability to be sensitive to the details of high spatial frequencies. In addition, the colour extension of Local Binary Pattern is not straightforward. In this article, we introduce the Colour Local Patterns (CLP) as a new vector texture feature able to characterise colour texture. The Colour Local Pattern inherits from the Local Binary Pattern construction and from the psycho-physical results, starting from the third Julesz’s conjecture. By defining the CLP in a perceptual colour space and by using a perceptual distance, we embed the notion of neighbourhood defined by Julesz and used in LBP. Then by applying a Fourier transform, we generate a signature vector for the local signature. The results achieved in classification tasks are higher around 10% in the rate of good classification in two databases with the largest number of images.