Total Variation And Its Usage In Image Processing

In this paper, we mainly process image by using total variation model. First of all, as to the inpainting problems in pixel domain, we propose a fast adaptive algorithm based on the TV model. It introduces different weights into the algorithm according to the features of image. Using this algorithm, better inpainting results can be achieved with much less computation time and then we make use of it in inpainting image in every direction instead of just confined to long and narrow blank. Besides, as to inpainting problem in wavelet domain, we introduce a new iterative model. In this model low frequency and high frequency will carry on processing separately, enormous speed up the image repair convergence rate, and even if as high as 90% wavelet coefficient have been lost, this model will also be able to have the very good repair effect. In the old model, to compute the curvature we have to transform the coefficients to the pixel domain, and then transform back to the coefficient domain for several times and this method can solve this problem and speed up the iterative convergence.Finally, we apply this method in image compression and image denoising. Experiments show that it can effectively eliminate the Gibbs oscillation in image compression and reduce the image information in empty area. At the same time, the TV model can also eliminate the"block effects"arising from wavelet denoising.