JPEG compressed image retrieval via statistical features
Introduction
At present, almost all digital images are stored in compressed formats, among which the format defined by joint picture expert group (JPEG) is widely used on Internet or image databases [1]. To do the feature detection or extraction for those compressed images, the conventional approaches need to decode the images to the pixel domain first, before carrying on with other existing image processing and analysis techniques [2]. This is not only time consuming, but also computationally expensive. Yet it becomes more and more important to improve the efficiency of indexing and retrieving compressed images. Therefore, a new wave of research efforts are directed to feature extraction in compressed domain [3], [4], [5]. As the inverse DCT (IDCT) is an embedded part of the JPEG decoder [1], and DCT itself is one of the best filters for the feature extraction, working in DCT domain directly remains to be the most promising area for compressed image processing and retrieval [6]. In addition, DCT also preserves a set of good properties such as energy compacting, and image data decorrelation. Thus, direct feature extraction from DCT domain could provide better solutions in characterizing the image content, apart from its advantage of eliminating any necessity of decomposing the image and detecting its features in pixel domain. Recently, Ngo et al. [5] developed an image indexing algorithm via reorganization of DCT coefficients in Mandala domain [7], and representation of color, shape and texture features in compressed domain. Their work demonstrated advantages in terms of indexing speed without significantly sacrificing the retrieval accuracy. However, their feature extraction techniques are significantly different from what we propose here. While their work is essentially highlighted by reorganization of those DCT coefficients, our work features in direct extraction of statistic parameters in DCT domain to combine the nature of texture and shape into an integrated feature for image indexing and retrieval. Further to the algorithm development, a fast JPEG image retrieval system is also designed and evaluated. The experimental results demonstrate that the presented system has good performance in terms of retrieval efficiency and effectiveness.
Section snippets
Computation of statistical parameters in DCT domain
In image processing or analysis, some statistical parameters are commonly employed to describe the image content. Among those statistical parameters, the moments or their functions are the most fundamental description of texture and shapes [8], [9], [10], [11]. Hu [8] is among the first group of people who derived results illustrating the algebraic invariant of two-dimensional moments. Following that, it becomes widely applied in feature detection and representation in image processing and
Image retrieval system design
According to JPEG compression standard, its full decompression process can be highlighted as: (a) entropy decoding via Huffman tables, (b) dequantization to obtain the DCT coefficients, and (c) IDCT to reconstruct blocks of pixels. For any content-based image retrieval (CBIR) systems that works in the pixel domain, this is a prerequisite operation or overhead operation before the key extraction and distance matching can be performed [2]. To improve the efficiency, we propose constructing keys
Performance evaluations
Along with the above design, a content-based image retrieval system (CBIR) is implemented by using C++. In general, there are two factors to evaluate the performance of a CBIR system [19]. One is the retrieval efficiency, which focuses on the speed of retrieval. The other is the retrieval effectiveness, which emphasizes the accuracy of the retrieval. As discussed and described in previous sections, our proposed algorithm does not need full decompression or full IDCT, and directly generates
Conclusions
In this paper, we have designed an image retrieval algorithm by directly and effectively computing the first and second moments from DCT domain. The relationship of the moments calculated inside various sized windows is also derived in DCT domain, which provides a complete freedom to extract statistical information from JPEG compressed images. As the proposed approaches are essentially operated in DCT domain, significant advantages can be exploited in saving computing cost and storage spaces
Acknowledgements
Finally, the authors wish to acknowledge the financial support from The Council for Museums, Archives & Libraries in the UK under the research Grant LIC/RE/082, and partial support from NSFC under Grant No. 60175031.
About the Author—GUOCAN FENG received his M.Sc. degree in Bio-mathematics from Zhongshan University, China, in 1988, and his Ph.D degree in Computer Science from Hong Kong Baptist University, Hong Kong, in 1999. He was a research fellow in School of Computing, University of Glamorgan, UK since October 2000 before joining University of Bradford in February 2002. He is currently an associate professor of Zhongshan University and a research fellow in School of Informatics, University of Bradford,
References (20)
- et al.
A critical evaluation of image and video indexing techniques in the compressed domain
Image Vision Comput.
(1999) - et al.
Recognition of JPEG compressed face images based on statistical methods
Image Vision Comput.
(2000) - et al.
Classification of partially occluded objects using 3-point matching and distance transformation
Pattern Recognition
(1994) - et al.
A fast content-based indexing and retrieval technique by the shape information in large database
J. Systems Software
(2001) The JPEG still picture compression standard
Commun. ACM
(1991)- B. Shen, Ishwar K. Sethi, Direct feature extraction from compressed images, Proceedings of the SPIE: Storage and...
- R. Reeve, K. Kubik, W. Osberger, Texture characterization of compressed aerial images using DCT coefficients,...
- et al.
Exploiting image indexing techniques in DCT domain
Pattern Recognition
(2001) - et al.
Pattern recognition experiments in the Mandala/Cosine domain
IEEE Trans. Pattern Anal. Mach. Intell.
(1983) Visual pattern recognition by moment invariants
IRE Trans. Inform. Theory
(1962)
Cited by (71)
Fast content access and retrieval of JPEG compressed images
2016, Signal Processing: Image CommunicationCitation Excerpt :The non-zero coefficients are also used to identify dominant color of compressed image and proposed a retrieval method based on the color descriptor [14]. Feng et al. [15] used statistics of the DCT coefficients say a set of moments, to form a feature vector to retrieve compressed images. Eom et al. [16] defined an edge histogram based on AC coefficients of DCT transform.
Image indexing and retrieval in JPEG compressed domain based on vector quantization
2013, Mathematical and Computer ModellingImage Segmentation with Multi-feature Fusion in Compressed Domain based on Region-Based Graph
2023, International Arab Journal of Information TechnologyEfficient histogram for region based image retrieval in thdiscrete cosine transform domain
2022, IAES International Journal of Artificial IntelligenceMachine learning approach for automatic diagnosis of Chlorosis in Vigna mungo leaves
2021, Multimedia Tools and ApplicationsAn image retrieval scheme based on block level hybrid dct-svd fused features
2021, Multimedia Tools and Applications
About the Author—GUOCAN FENG received his M.Sc. degree in Bio-mathematics from Zhongshan University, China, in 1988, and his Ph.D degree in Computer Science from Hong Kong Baptist University, Hong Kong, in 1999. He was a research fellow in School of Computing, University of Glamorgan, UK since October 2000 before joining University of Bradford in February 2002. He is currently an associate professor of Zhongshan University and a research fellow in School of Informatics, University of Bradford, UK. His current research interests include digital image processing, pattern recognition, human face recognition and image retrieval, indexing in compressed domain.
About the Author—JIANMIN JIANG received B.Sc. degree from Shandong Mining Institute, China, in 1982, M.Sc. degree from China University of Mining and Technology in 1984, and Ph.D. from the University of Nottingham, UK, in 1994. From 1985 to 1989, he was a lecturer at Jiangxi University of Technology, China. In 1989, he joined Loughborough University, UK, as a visiting scholar and later moved to the University of Nottingham as a research fellow. In 1992, he was appointed a lecturer of electronics at Bolton Institute, UK, and moved back to Loughborough University in 1995 as a lecturer of computer science. From 1997 to 2001, he worked as a professor of Digital Imaging and Data Compression at the School of Computing, University of Glamorgan, Wales, UK. In 2002, he joined the Department of Electronic Imaging & Media Communications, University of Bradford, as a Professor of Digital Media. He is a chartered engineer, fellow of RSA, and a consulting professor at Chinese Academy of Sciences. His research interests include information retrieval, image processing in compressed domain, data compression, digital video coding, stereo image coding, medical imaging, computer graphics and neural network applications. He has published over 130 refereed research papers and invented one European patent (EP01306129) filed by British Telecom Research Lab on pixel extraction from compressed videos without decompression.