Image Retrieval using Multi Texton Co-occurrence Descriptor and Discrete Wavelet Transform

This paper describes an efficient algorithm for Content-Based Image Retrieval (CBIR) based on Multi Texton Co-Occurrence Descriptor (MTCD) and Haar (Wavelet) namely MTCD-H. The problems from the previous research were the computational speed and the low value of precision. The data that was used consisted of 10000 Corel images and 300 batik images. MTCD used the RGB colour feature, the Sobel edge detection, and global feature using Gray Level Co-Occurrence Descriptor (GLCM). Wavelet was considered an approach that could increase the precision value as well as reduce the features. This paper combined the MTCD with Haar for the image extraction process in order to increase the computational speed and the precision value. The contribution of this paper was aiming for extracting the Wavelet feature on a grayscale image before extracting the feature using GLCM. The results showed an increase of precision value pointed at 3.36 for batik images and 5.11 for the Corel images. In addition, the computational speed for batik images was performed 84. 35s faster as for the Corel images 2988 faster. Based on the specified results, it could be concluded that the MTCD-H were effective in reducing the computational speed as well as increase the precision.