Wavelet Transform-based Image Denoising And Compression Algorithm

The digital image is often corrupted by noise in its acquisition and transmission, which has a negative effect on the subsequent image processing tasks such as objects segmentation and identification, so removing noise is a significant image preprocessing task. On the other hand, with the rapid development of digital communications, digital image data have an exponential growth; image compression has become one topic of the most active research in image processing field. Wavelet transform has excellent time-frequency localization capability and good decorrelation ability and it can effectively extract information from signal, so that Wavelet transform is widely applied in image processing.A main research on image de-noising and compression algorithm based on wavelet transform is presented in this paper. And this paper proposes the modified wavelet threshold denoising algorithm and the modified set partitioning in hierarchical trees(SPIHT) algorithm. Wavelet threshold denoising algorithm is one of the widest researches. A lot of modified wavelet threshold denoising algorithms have been proposed. However, those algorithms neglect the denoising in the low frequency subband. Experimental results show that the low frequency subband still has a small amount of noise, for this reason adaptive median filter is utilized in the low frequency subband to further remove the noisy coefficients. And the Universal threshold de-noising algorithm is employed in the high frequency subbands.All of the wavelet coefficients are encoded simultaneously by the traditional SPIHT algorithm when the threshold is relatively large, so that a large quantity of redundant 0 bits are generated, which affects the efficiency of coding algorithm. This paper presents a modified SPIHT that is based on the aforementioned problem. First, the proposed improvement judges if there are significant coefficients in the list of insignificant sets(LIS) for the current threshold instead of encoding the sets in LIS directly. If all the coefficients in LIS are insignificant, the modified SPIHT will skip the coding of LIS. On the contrary, LIS is encoded as the original SPIHT. Second, a flat bit is added to the encoding process of the list of insignificant pixels(LIP) to test whether all the significant coefficients have been encoded. The improvement of LIP will skip the encoding of LIP once all the significant coefficients have been encoded.It can be observed from the experimental results that the modified wavelet threshold denoising algorithm gains the higher Peak Signal to Noise Ratio(PSNR) of the reconstructed image for the lager noise variance(the more serious the noise pollution) compared to the raw method. However the PSNR of the reconstructed image goes done slightly as the noise variance decreases. Furthermore, compared with the raw method, the modified SPIHT achieves higher quality of the reconstructed image at the same compression ratio compared to the raw method. Especially at low bit rates, the improvement of PSNR value is particularly pronounced.