Content-Based Image Processing Techniques Research In Compressed Domain

Traditional image processing techniques are used mainly in pixel domain, for most compressed images, we cannot process them until they have been decompressed, which will affect system's time-real performance and agility. Therefore, compressed-domain-based image processing techniques become the investigation interests in recent years. Discrete cosine transform (DCT) technique was used in image compression early, most compressed-domain-based image processing techniques are studied in the DCT domain; Wavelet transform technique has time-frequency localization character and multi-resolution character, which is used in image coding domain widely, so it is significative to study the wavelet-domain-based image processing techniques. Aimed at the international image coding standards, the dissertation studies the image processing techniques in DCT domain and wavelet transform domain.Above all, dissertation illuminates compressed-domain-based image processing techniques' basic idea, as well as the research method, and it summarizes the expressions of image feature. JPEG and JPEG2000 are dominating image coding standards, dissertation discusses the compression algorithms used in the two standards in detail, and illuminates the flow of image coding.Afterward, dissertation does the research of DCT-domain-based image processing techniques. After analyzing Discrete Cosine Transform and the coding characteristics of JPEG, paper investigates DCT block's data distribution, on the basis of data distribution characters, we propose a fast algorithm for contour extraction using DCT blocks' coefficients, the algorithm firstly extracts each DCT block's DC coefficient and two AC coefficients, and secondly organizes them, according to the coefficients' statistical feature, we figure out a threshold to extract a loose contour. In the post-processing, we introduce curve fitting technique to achieve image's close contour, emulational result indicates that the algorithm can extract image's contour without full decompression.At last, dissertation studies the wavelet-transform-domain-based image processing techniques. Paper detailedly analyses and summarizes the data feature of image in wavelet transform domain, proposes a target extraction technique using high-frequency coefficients, the method utilizes mathematical morphology, it deals little data simply, emulational result indicates that this algorithm can extract target efficiently and practicably. The second aspect is aimed at character location. Dissertation combines current character location techniques, and proposes a wavelet-transform-domain-based algorithm, the algorithm utilizes '米' structure to do weighted processing of wavelet coefficients, then uses region-based Pixel-Clustering algorithm to differentiate character region from non-character region, finally we introduce projective histogram to locate the character region, emulational result indicates that this method can detect character region accurately and effectively, It also is robust in multi-scale character location and anti-noise.