Picture Inpainting
-
Python implementation of the paper IMAGE INPAINTING VIA SPARSE REPRESENTATION, ICASSP 2009
-
Picture inpainting using Python Package Dictlearn.
Paper based method
Run
To run the reimplemented project for the paper, place your original image and corresponding mask in the path ./pictures like this:
-pictures
-image file
-image_mask file
And then run:
python run_paper.pyOptions
--picture_path : path to the original picture
--mask_path : path to the mask, it has to be black and white, the black part indicates the missing pixels
--ave_dir : path to save the results
--patch_size : side-length for a square patch
--step : step interval to fetch the patches
--alpha : alpha value for Lasso
--max_iter : max_iteration time for Lasso
--tolerance : tolerance value for Lasso
--local : whether to build the dictionary locally
Results
The results will be save at ./results,including the inpainted image and a report for a single experiment will be save at ./results/results.txt. An example is like below:
>>> Experiment Time: 20190610-193633
>>> Experiment Settings:
patch_size: 101 | step: 25 | alpha: 0.001000 | tolerance: 0.000100 | max_iter: 10000 | local: 1
>>>Experimental Attribute:
total patch num: 470 | total missing pixel num: 11414 | average iteration: 100 | total time used: 293 s
>>>Experimental Metrics:
MSE: 0.002956 | PSNR: 25.293428 | SSIM: 0.957037 >>> Inpainted picture save at:
./results/hill_inpaint_20190610-193633.jpg Dictlearn based method
Run
To run the dictlearn based method, place your original image and corresponding mask in the path ./pictures the same as above:
-pictures
-image file
-image_mask file
And then run:
python run_package.pyResults
The results will be save at ./results/package/
Should you have any advice on the project or any problem using it, feel free to let me know. Issues are welcome!