Color and texture features extraction on content-based image retrieval

Abstract

The study on Content Based Image Retrieval (CBIR) has been a concern for many researchers. To conduct CBIR study, some essential things should be considered that are determining the image dataset, extraction method, and image measurement method. In this study, the dataset used is the Oxford Flower 17 dataset. The feature extraction employed is the feature extraction of the HSV color, the Gray Level Cooccurrence Matrix (GLCM) texture extraction feature, and the combination of both features. This study is purposely generates precision from CBIR test based on the proposed method. At first, digital image is segmented by applying thresholding. Moreover, the image is converted into vector to be subsequently processed using feature extraction. Further, the similarity level of the image is measured by Euclidean Distance. Tests on the system are based on segmented and unsegmented image. The system test with segmented image yields mean average precision of 83.35% for HSV feature extraction, 83.4% for GLCM feature extraction, and 80.94% for combined feature extraction. Meanwhile, the system test for unsegmented image generates mean average precision of 82.64% for HSV feature extraction, 87.32% for GLCM feature extraction, and 85.73% for extraction of combined features.