CONCEALMENT OF TRANSMISSION ERRORS IN JPEG-2000 IMAGES USING ADAPTIVE LINEAR PREDICTION AND CROSSBAND INFORMATION

• Main Authors: Shih-Hsiang-Lin, 林士湘
• Other Authors: Jin-Jang Leou
• Format: Others
• Language: en_US
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/73920602644829208137

碩士 === 國立中正大學 === 資訊工程研究所 === 88 === ABSTRACT In this study, the error concealment approach (using adaptive linear prediction and crossband information) to transmission errors in JPEG-2000 images is proposed. For entropy-coded JPEG-2000 images, a transmission error in a codeword may cause the underlining codeword and its subsequent codewords to be misinterpreted, resulting in a great degradation of the received images. Here a transmission error may be a single-bit error or a burst error containing N successive error bits. The objective is to conceal transmission errors in JPEG-2000 images, i.e., to recover high-quality JPEG-2000 images from their corresponding corrupted images by utilizing the information of the correctly-received blocks and the crossband characteristics within JPEG-2000 images. Because the human visual system (HVS) is more sensitive to error corruption in low frequency regions and blocks in low frequency subbands convey much more energy (information) than those in the higher frequency subbands, loss of even one block in a low frequency subband will cause severe degradation in image quality, whereas loss of one block in a high frequency subband will cause relatively slight degradation in image quality. Therefore, when JPEG-2000 images are decomposed into 5 wavelet levels (default), for the higher frequency subbands in the 4th and 5th levels, after applying the error detection procedure, the subband samples in the corrupted blocks are simply replaced by zeroes. Adaptive linear prediction and crossband information are performed only on the 3 lower levels, or equivalently the 7 lower frequency subbands. Within the 3 lower levels, because the correlations in different subbands in horizontal and vertical directions are totally different, depending on (1) the position of the corrupted block (in wavelet transform domain), (2) the condition of correctly-received blocks in neighboring subbands, and (3) the statistical property (variance) of the corrupt block, the proposed approach adaptively selects one among 225 sets of linear prediction coefficients to conceal a corrupted block in wavelet transform domain. Within a corrupted block, depending on a subband sample is either an edge subband sample or a nonedge subband sample, different concealment strategies are employed to conceal the subband sample. Based on the simulation results obtained in this study, the proposed approach can recover high-quality JPEG-2000 images from the corresponding corrupted JPEG-2000 images. This shows the feasibility of the proposed approach.