Content Based Image Retrieval using Color Histogram and Discrete Cosine Transform

摘要

This paper proposed a color image retrieval approach based on images' content. This approach is based on extracting an efficient combination of low visual features in the image; color and texture. To extract the color feature, color histogram was used, where the RGB color space was converted into HSV color space, then the color histogram of each space was taken. To extract the texture feature, DCT transformation was used, and DC coefficients are taken meanwhile neglecting AC coefficients. The experimental results were analyzed on the basis of three similarity measures, Manhattan Distance , Euclidean Distance and Mean Square Error. MD similarity measure proved its efficiency in retrieval process compared with other similarity measures at both the execution time and retrieval accuracy. The accuracy and efficiency of the system were evaluated using the precision and recall metrics. The results obtained from the proposed approach showed good results when considering precision measure in evaluation process. The precision was increased by (8.3%) rate compared to the best result of previous studies.