A new feature-based wavelet completed local ternary pattern (Feat-WCLTP) for texture image classification

• Main Authors: Shamaileh, Abeer, Rassem, Taha H., Siau Chuin, Liew, Al Sayaydeh, Osama Nayel
• Format: Article
• Language: English; English
• Published: IEEE 2020
• Subjects: QA75 Electronic computers. Computer science; QA76 Computer software
• Online Access: http://umpir.ump.edu.my/id/eprint/28453/1/A%20New%20Feature-Based%20Wavelet%20Completed.pdf; http://umpir.ump.edu.my/id/eprint/28453/2/A%20New%20Feature-Based%20Wavelet%20Completed%20Local%20Ternary%20Pattern.pdf; http://umpir.ump.edu.my/id/eprint/28453/; https://doi.org/10.1109/ACCESS.2020.2972151; https://doi.org/10.1109/ACCESS.2020.2972151

LBP is one of the simplest yet most powerful feature extraction descriptors. Many descriptors based on LBP have been proposed to improve its performance. Completed Local Ternary Pattern (CLTP) is one of the important LBP variants that was proposed to overcome LBP's drawbacks. However, despite the impressive performance of CLTP, it suffers from some limitations, such as high dimensionality, thereby leading to higher computation time and may affect the classification accuracy. In this paper, a new rotation invariant texture descriptor (Feat-WCLTP) is proposed. In the proposed Feat-WCLTP descriptor, first the redundant discrete wavelet transform RDWT is integrated with the original CLTP. Then, CLTP is extracted based on the LL wavelet coefficients. Next, the mean and variance features are used to describe the magnitude information instead of using P-dimensional features as the normal magnitude components of CLTP. Reducing the number of extracted features positively affected the computational complexity of the descriptor and the dimensionality of the resultant histogram. The proposed Feat-WCLTP is evaluated using four texture datasets and compared with some well-known descriptors. The experimental results show that Feat-WCLTP outperformed the other descriptors in terms of classification accuracy. It achieves 99.66% in OuTex, 96.89% in CUReT, 95.23% in UIUC and 99.92% in the Kylberg dataset. The experimental results showed that the Feat-WCLTP not only overcomes the CLTP's dimensionality problem but also further improves the classification accuracy.