LOW LEVEL FEATURES FOR IMAGE INDEXING AND RETRIEVAL IN THE COMPRESSED DOMAIN

Abstract

Abstract— in this paper, the work used to explore three of low level feature descriptors, namely: binary texture, energy-based, orientation based content, and DC descriptors. A simple binary algorithm is applied to transform DCT values into binary, a histogram is generated by using horizontal and vertical features, and energy based of average DCT values to construct indexing key for image retrieval.  The experiment shows that binary texture descriptor demonstrates the highest in average precision amongst other low level descriptors.  The research was also explored to calculate the effectiveness of image retrieval in DCT and by using commercial image processing package Adobe Photoshop Album (APA) as comparison.