Content-based Face Color Image Retrieval using Multi-feature Fusion Extractor Methods

The retrieval of visual image content has been the most active research in various applications. In this paper, the benchmark datasets have been used as a fast screening process for extracting representative facial features. Despite extracting relevant features information from the entire face, local features focused on the segmented regional area have proved to be more effective as suggested in the literature search. In doing so, four labeled region area was chosen before it was combined together to create a new sample image as input data for further analysis. To enhance the image representation, variational color spaces conversion are used and the first four color moments are selected for acquiring color information about the image. Also, the main five texture features are concatenated later with the color moments to analyze the complementary effects of color features in texture. In total, the nine selections of feature fusion methods have been presented, whereas the high dimensional space has been through the dimensional reduction process. The experimental result demonstrates that higher image content retrieval accuracy can be obtained by applying the CM+BSIF feature for YCbCr thermal image (0.4688 ± 0.1481) and CM+BSIF+Tamura for HSV visible image (0.4631 ± 0.1512).