Ask for advice on the recurrence of Figure 27

[GY5134]

Dear professor, hello. Thank you for providing the code, I am downloading your code, set lambda=30;iter=15;gamma=10/255; to run your image with fish according to the ILS_Welsch function, the result is a lot of patches, it is not you at all The results appear in the article, so I ask you. If it is convenient for you, please help me.

[wliusjtu]

Dear professor, hello. Thank you for providing the code, I am downloading your code, set lambda=30;iter=15;gamma=10/255; to run your image with fish according to the ILS_Welsch function, the result is a lot of patches, it is not you at all The results appear in the article, so I ask you. If it is convenient for you, please help me.

Many thanks for your interests in our paper. In fact, since the Welsch penalty function seldom penalizes very large gradients and textures can sometimes have very high contrast, we pre-smooth the input image with a Gaussian fiter of a small variation to properly reduce the contrast of small structures. We have indicated this in the beginning of the third paragraph of Sec. 5.1. For the illustrated image, we pre-smooth it with the a bilatera filter (dowloaded at http://people.csail.mit.edu/jiawen/software/bilateralFilter.m) with sigma_s=0.75, sigma_r=100(sufficient large value of sigma_r to simulate the Gaussian filter). In addition, there is a typo in our paper, the value of lambda shouble be lambda=3 instead of 30. Please also pay attention to the image size, different image sizes may require a Gaussian filter of different values of sigma_s. Hope my reply may be useful to you.