A NOVEL EFFICIENT HISTOGRAM EXTRACTION SCHEME FOR CONTENTS BASED IMAGE PROCESSING IN DCT DOMAIN Minyoung Eom and Yoonsik Choe Dep. of Electrical and Electronics Engineering, Yonsei University 134 Shinchon-dong, Seodeamoon-goo, Seoul, 120-749, Korea Emy0606@yonsei.ac.kr yschoe@yonsei.ac.kr ABSTRACT Due to development of internet network environments and data compression techniques, the size and amount of multimedia data has greatly increased. They are compressed before transmission or storage. Dealing with these compressed data such as video retrieval or indexing requires decompression procedure most of the time. This causes additional computations and increases the processing time. In video retrieval and indexing a histogram is one of the most frequently used tools. Efficiency of extracting such histograms will drop down if decompression gets involved. We propose a novel scheme for extracting histograms from images that are transformed into the compressed domain by 8x8 DCT(Discrete Cosine Transform). In this scheme an averaged version of original image is obtained by a simple linear combination of DCT coefficients with the sets of coefficients we designed. KEY WORDS Color Histogram, Edge Histogram, image processing, DCT Domain 1. Introduction Due to rapid development of data communication techniques and network environment, the use of multimedia data in various applications such as internet, satellite broadcasting, mobile communication and data retrieval is constantly increasing. Such multimedia data is compressed with various algorithms, JPEG, JPEG2000, MPEG-1/2/4 and H.261/263/264, for the efficient storage, management and transmission. But the compressed data are transformed back into the spatial domain when they are used in various applications [1]. So the efficiency of using multimedia data is decreased because of the additional computation and processing time for the decompression procedure. A histogram is one of the most frequently used tools in manipulating multimedia data. Up to now, the extraction of the histogram has been mostly conducted in spatial domain owing to its accuracy. However, the extraction processes including the decoding procedure are time consuming as well as computationally expensive. So some researchers have worked to directly extract an image in DCT domain. G. C. Feng and J. Jiang proposed a scheme that reconstructs an average version image of the original one from 8x8 DCT coefficients without taking IDCT(Inverse DCT). Although the color histogram extracted from the reconstructed image by this scheme is similar to the one from the original image, this scheme has a limit that it can only reconstruct an image of quartered resolution [2] and has some problems in terms of accuracy due to the approximation. In [3], we were proposed an edge histogram extraction scheme in the compressed domain. The histogram is extracted by the ratio of tow AC coefficients using the characteristics of them. In spite of the computational efficiency of this scheme, it is limited to the extraction of edge histograms. In this paper, we propose a novel scheme that directly extracts an image from the DCT coefficients without full decoding procedure using a simple linear combination of DCT coefficients. Using this scheme, we can extract not only the quarter-resolution image but also the half- resolution one, which give us histograms which are almost similar to the histograms of the original image. In addition, this scheme is not limited to the extraction of a color histogram or an edge histogram. This paper is organized as follows. In section 2, the existing method is reviewed, and then present our scheme in section 3 in detail. Histograms extracted from the original image and the extracted images by both the former and our schemes are represented and compared in section 4. Finally, section 5 concludes this paper. 2. Image Extraction in DCT domain (IED) G. C. Feng and J. Jiang proposed IED which extracts an average version of an original image from the 8x8 DCT coefficients without performing IDCT. For an 8x8 DCT coefficient block shown in Fig. 2, this scheme calculates the four values m , , and as shown in Fig. 1. Each m is the average value of the pixels in a 4x4 block. Then they are approximately represented by linear combinations of the four DC and AC coefficients, c , , c and c , in 8x8 DCT block as below Equation (1) ~ (4) 11 12 m 21 m 22 m ij 00 01 c 10 11 554-080 164