Anisotropic Diffusion-Based Color Texture Analysis for Industrial Application

Anisotropic Diffusion-Based Color Texture Analysis for Industrial Application

Rohini A. Bhusnurmath (Karnataka State Akkamahadevi Women's University, India) and Prakash S. Hiremath (KLE Technological University, India)
DOI: 10.4018/978-1-7998-2736-8.ch002
OnDemand PDF Download:
No Current Special Offers


This chapter proposes the framework for computer vision algorithm for industrial application. The proposed framework uses wavelet transform to obtain the multiresolution images. Anisotropic diffusion is employed to obtain the texture component. Various feature sets and their combinations are considered obtained from texture component. Linear discriminant analysis is employed to get the distinguished features. The k-NN classifier is used for classification. The proposed method is experimented on benchmark datasets for texture classification. Further, the method is extended to exploration of different color spaces for finding reference standard. The thrust area of industrial applications for machine intelligence in computer vision is considered. The industrial datasets, namely, MondialMarmi dataset for granite tiles and Parquet dataset for wood textures are experimented. It was observed that the combination of features performs better in YCbCr and HSV color spaces for MondialMarmi and Parquet datasets as compared to the other methods in literature.
Chapter Preview


Computer vision task in artificial intelligence aims in recognizing the patterns in images. The focus is to identify patterns in images based on textural information. Applications of texture analysis in object recognition in various image processing modalities have made the task of efficient texture feature extraction more imperative and challenging. There is a definite need to develop novel algorithms which are faster and more accurate in texture representation and classification. In this direction, some authors have investigated the diffusion model based texture analysis in an attempt to address the challenges in texture classification. There is a constant need to increase recognition accuracy. Further, the traditional techniques of machine learning often require human expertise and knowledge to develop features tailor made for a particular application. Thus, the success of artificial intelligence lies in developing strong features rather than just machine learning algorithms that implement them. Therefore, the intelligent machine learning techniques do need robust textural features for accurate pattern analysis.

Texture analysis has been a prominent aspect in computer vision and image processing. A survey of texture analysis is given in (Tuceryan & Jain, 1998; Pietikainen, 2000; Petrou & Sevilla, 2006; Xie, 2008; Hermanson & Wiedenhoeft, 2011; Ahmadvand & Daliri, 2016a, Heurtier, 2019). The chrominance information is also incorporated into texture features in (Paschos, 2000; Mirmehdi & Petrou, 2000; Bombardier & Schmitt, 2010; Bianconi, et al., 2013; Hiremath & Bhusnurmath, 2014c; Hoang, 2018; Porebski, et al., 2018). A number of feature learning methods for texture classification based on statistical features have been proposed in recent years (Lu, et al., 2015a-2015d; Doost & Arimani, 2013; Zhu & Wang, 2012). It is observed that there is no universal set of textural features that serve the purpose of a recognition method for different tasks. In image analysis, texture recognition techniques deal with various textures present in the images. These texture measures must be robust and invariant to texture structures. The low computational complexity is necessary for any real time application of the methods.

Texture classification can generally be used as an idea to solve many real world problems. Many computer vision algorithms today use the idea of texture classification to accomplish the task. These pattern recognition algorithms often view the subject of interest as different textures, and classify those accordingly using texture analysis techniques.

Grading is the problem of automatic recognition of products which has wide applications in industrial products. The examples of such products include, textile (Sheriff et al., 2018), ceramic tiles (Fernandez et al. 2011; Ferreira & Giraldi 2017), leather (Murinto et al., 2018; Liong et al.,2019). In the industries such as parquet (Bianconi, et al., 2013; Bello-Cerezo, et al., 2019), wood (Jahanbanifard et al., 2019; Hiremath, & Bhusnurmath, 2016c, 2016d, 2017b; Bhusnurmath & Hiremath, 2019; Shustrov, et al., 2019), natural stone (Bianconi, et al., 2015), etc.

Texture classification techniques are popularly used in industrial applications in grading products based on visual appearance (CIE 2006; Eugene, 2008). These systems employ image processing techniques with the help of two main visual features: texture and colour (CIE, 2006). The examples of a computer vision application in the areas of industrial automation that involve texture classification are wood classification and granite tiles classification which is done manually and is subjective. Hence, there is an emergent need to perform this task automatically.

Complete Chapter List

Search this Book: