Article Preview
TopIntroduction
Texture classification problem is typically encountered in solving many real-world problems related to computer vision for object recognition based on its surface properties. The pattern recognition algorithms often view the objects of interest as different textures, which are classified using texture analysis techniques (Pietikainen, 2000). Automatic classification of a product into lots of similar visual appearance, a problem sometimes referred to as grading, has found interesting applications in many industrial products, such as paper (Turtinen et al., 2006; Maldonado & Grana, 2009), ceramic tiles (Boukouvalas et al., 2000; Kukkonen et al., 2001; Jiaoyan & Xuefeng, 2004) and leather (Yeh & Perng, 2001). Granite and wood industries are also concerned with the development of automatic machine vision system for grading. The manual quality control procedures adopted are time-consuming, subjective and non-repetitive.
Cusano et al. (2016) have investigated the effects of varying light conditions on texture features, with and without normalization. Bianconi et al. (2017) have introduced the improved opponent colour local binary pattern (LBP) descriptor for colour texture classification and experimented it on eight colour datasets, which is observed to perform better than other LBP variants. Ledoux et al. (2017) have introduced LBP extension for color texture descriptors and found that the combination of two color orders of small size performs better for classification problem. A number of effective texture analysis methods have been reported in the literature (Hiremath & Shivashankar, 2008; Hiremath & Bhusnurmath, 2014; Hiremath & Bhusnurmath, 2016a). Linear discriminant analysis (LDA) is found to be efficiently used in pattern recognition (Shivashankar & Hiremath, 2011). Perona and Malik (1990) introduced anisotropic diffusion for image smoothing while keeping the edges sharp in texture analysis. Multiresolution LDBP descriptors using anisotropic diffusion method is explored to achieve better classification results with low computational costs (Hiremath & Bhusnurmath, 2017). A number of methods for automatic classification of granite textures have been proposed in the past few years (Bianconi et al., 2009; Bianconi et al., 2012; Gonzalez et al., 2013). LBP based methods and its variant have been effectively used for granite texture classification (Fernandez et al., 2011; Fernandez et al., 2013; Bianconi et al., 2015; Paci et al., 2013). Kylberg and Sintorn (2013) have used local binary pattern approach to texture analysis of granite tiles and investigated noise robustness. More recently, Bianconi et al. (2013) proposed method for sorting of hard wood parquet slabs based on various texture descriptors in different colour spaces.
The digital image processing methods work on images captured by industrial cameras, which are usually RGB images. These RGB images can be either used ‘as is’ or converted into different colour spaces, such as HSV, CIE Lab, etc. In both cases, the aim is to extract global statistical descriptors that characterize the colour content of the images. In the design of an expert system for wood and granite tiles grading, that use image processing techniques, one has to deal with the choice of the right kind of colour space and the appropriate texture descriptor.