An efficient texture classification algorithm using integrated Discrete Wavelet Transform and local binary pattern features. (December 2018)

• Record Type: Journal Article
• Authors: Gopala Krishnan, K.; Vanathi, P.T.
• Is Part Of: Cognitive systems research. Volume 52(2018)
• Page Start: 267
• Page End: 274
• Publication Date: 2018-12
• Subjects: Texture classification -- Local binary pattern -- Discrete Wavelet Transform -- k-NN classifier; Cognition -- Periodicals; Cognitive engineering (System design) -- Periodicals; Artificial intelligence -- Periodicals; 153.05
• DOI: 10.1016/j.cogsys.2018.07.015
• Languages: English
• ISSNs: 1389-0417

Abstract

Abstract: This manuscript is keen to the Texture Classification problem. Texture is mainly defined as measuring the variation in the surface intensity such as regularity, smoothness, coarseness, etc. Texture classification is one of the most important issues in image processing and computer vision. Orientation, scale, image transitions or singularities such as edges, and the other visual appearance are the major problems in texture classification. Already works have done in texture classification by using Discrete Wavelet Transforms (DWT) and Local Binary Pattern (LBP) separately. The above techniques give minimum classification Accuracy. LBP is considered as an effective method but its performance is lower if the image has poor quality. We propose a technique to characterize the texture properties based on DWT using Local Binary Pattern. In this proposed work, input texture images are decomposed using single level Discrete Wavelet Transform. Then LBP features are extracted from all sub bands. The extracted LBP features for sub bands are combined to form main feature (1024 features). Image classification is done by using k-Nearest Neighbour (kNN) Classifier. The experiments validation are achieved by using four standard data sets (KTH-TIPS, KTH-TIPS-2a, Brodatz and Curet). The results are compared with Dense Micro block Difference (DMD) feature descriptors. The experimental result shows that the proposed method outperforms than the existing techniques. Also reduce the computational complexity and minimum computational time than the existing classification techniques.