Efficient texture image retrieval of improved completed robust local binary pattern

Improved Completed Robust Local Binary Pattern is one of the robust texture extraction for image retrieval that rotation invariant (ICRLBP). ICRLBP has proven that can increase the precision, recall, and computation time from its previous work by 21.14%, 20.03%, and 56 times, respectively, on four different texture image dataset. ICRLBP, however, has a lot of feature, thus require more time during recognition process. Moreover, it leads to high time consuming and curse of dimensionality. To overcome those issues, in this paper, we try to reduce insignificant or unnecessary ICRLBP attributes and examine the effect of reducing number of attributes on precision and recall of the retrieving images. The methods we used to reduce the ICRLBP attributes are Correlation-based Feature Selection (CFS) and Pearson's-based Correlation. The experiment results show that those feature selections not only can reduce number of attributes but also can improve precision, recall, and computation time. For S_M_C feature (sign, magnitude, and center features of ICRLBP are ploted on histogram jointly), CFS can reduce up to 95% number of attributes and improve precision, recall, computation time up to 7.5%, 7.1%, 11.42 times, respectively. For M_C feature (magnitude and center features of ICRLBP are ploted on histogram jointly), CFS can reduce up to 4.2% number of attributes and improve precision, recall, computation time up to 4.2%, 4%, 1.1 times, respectively. It indicated that Correlation-based Feature Selection (CFS) and Pearson's-based Correlation can reduce the ICRLBP attributes effectively.