Experimentation on the Use of Chromaticity Features, Local Binary Pattern, and Discrete Cosine Transform in Colour Texture Analysis

Abstract

Abstract. This paper describes a method for colour texture analysis, which performs segmentation based on colour and texture information. The main goal of this approach is to examine the contribution of chromaticity features in the analysis of texture. Local binary pattern and discrete cosine transform are the techniques utilised as a tool to perform feature extraction. Segmentation is carried out based on an unsupervised texture segmentation method. The performance of the method is evaluated using different chromaticity features and also using the ROC curves. The results indicate that the inclusion of colour information improves the segmentation performance.

Local binary patterns versus signal processing texture analysis: a study from a performance evaluation perspective

“…According to these studies, texture can be conceptualised as a statistical or geometric repetition of primitive descriptors (micro patterns) in the image and specific measures such as roughness, regularity, linearity, frequency, directionality, granularity and density can be employed to attain texture discrimination (one prominent example is the theory of textons that has been proposed by Julesz (1981) in the early 1980s). While the surfaces of the imaged objects are often defined by an unbounded variety of textures, the task relating to the identification of the optimal texture analysis approach proved extremely challenging and the substantial efforts devoted by the vision community in the field of texture analysis were justified, as the availability of a robust texture descriptor will be extremely beneficial for a large spectrum of applications (Manjunath and Ma, 1996;Kovalev et al, 2001;Mäenpää et al, 2003;Nammalwar et al, 2003;Ghita et al, 2005;Rodriguez and Marcel, 2006;Xie and Mirmehdi, 2007;Tosun and Gunduz-Demir, 2011). Due to its intrinsic complexity, this fundamental image property has been researched for a number of decades and there is a large degree of consensus among vision researchers that texture analysis can be divided into four major categories: statistical, model-based, signal processing and structural, with statistical and signal processing techniques being the most investigated.…”

CTex—An Adaptive Unsupervised Segmentation Algorithm Based on Color-Texture Coherence

“…The texture information is extracted by applying a local linear transform, while the colour is sampled by the moments calculated from the colour histogram. This approach is extremely appealing, since the contribution of colour and texture can be easily quantified in the segmentation process and it has been adopted by a large number of researchers [20]- [22]. Another related implementation has been proposed by Carson et al [23].…”

Colour Texture Analysis

“…Although the motivation to use colour and texture information jointly in the segmentation process is clear, how best to combine these features in a colour-texture mathematical descriptor is still an open issue. To address this problem a number of researchers augmented the textural features with statistical chrominance features 25,52,53 . Although simple, this approach produced far superior results than texture only algorithms and in addition the extra computational cost required by the calculation of colour features is negligible when compared with the computational overhead associated with the extraction of textural features.…”

“…The first step of the algorithm recursively splits the image hierarchically into four sub-blocks using the texture information extracted using the Local Binary Patterns/Contrast (LBP/C) method 53,56,57 . The splitting decision evaluates the uniformity factor of the region under analysis that is sampled using the Kolmogorov-Smirnov Metric (MKS).…”