ETH-Image-Inpainting-master
所属分类:图形图像处理
开发工具:matlab
文件大小:4705KB
下载次数:5
上传日期:2018-04-10 09:08:52
上 传 者:
markyT
说明: 这个文件描述了我们用来获取结果的MATLAB文件,
以及如何再现它们的说明。文件列表包括:buildDictionary.m dictionary.mat get_degrees inPainting.m
OMP.m overDCTdict.m overlap_col2im.m overlap_im2col.m peel_mask.m SMP.m
(This file describes matlab files that we used to obtain our results,
together with instructions for how can they be reproduced.)
文件列表:
code\buildDictionary.m (351, 2018-03-05)
code\code.zip (952968, 2016-02-15)
code\dictionary.mat (942035, 2016-02-15)
code\EvaluateInpainting.m (1207, 2016-02-15)
code\get_degrees.m (1021, 2018-03-08)
code\inPainting.m (1120, 2018-03-08)
code\OMP.m (1913, 2018-03-09)
code\overDCTdict.m (415, 2018-03-08)
code\overlap_col2im.m (2328, 2018-03-08)
code\overlap_im2col.m (674, 2016-02-15)
code\peel_mask.m (1389, 2018-03-08)
code\runtime.m (40, 2016-02-15)
code\SMP.m (1697, 2016-02-15)
report.pdf (2923356, 2016-02-15)
code (0, 2018-03-08)
#Image Inpainting Project: ETH Computational Intelligence Lab 2015
This file describes matlab files that we used to obtain our results,
together with instructions for how can they be reproduced.
I. File List
------------
* buildDictionary.m: Returns a prebuild matrix 'dictionary.mat' or builds a twice overcomplete DCT of supplied dimension if such matrix is not found
* dictionary.mat: Our dictionary obtained with K-SVD algorithm
* get_degrees: Takes image mask M in 2-D image form and counts for each pixel the number of its neighbours in mask
* inPainting.m: Perform the actual inpainting of the image
* OMP.m: Our implementation of the Orthogonal Matching Pursuit
* overDCTdict.m: Computes overcomplete Discrete Cosine Transform of supplied dimensions
* overlap_col2im.m: Recombines overlapping (d x d) patches into an image reconstruction weighting them by their signal-to-noise ration
* overlap_im2col.m: Extracts (d x d) patches from image I after every 'overlap' pixels
* peel_mask.m: Removes boundary pixel layer from the mask
* SMP.m: Performs sequential mask peeling inpainting algorithm, as described in our paper
II. Reproducing results
-----------------------
Anyone wishing to reproduce our results should use the default parameter values specified in the files. To reconstruct a single channel image, one needs to perform the following. Represent the grayscale image as a matrix I in [0,1] range and get a binary mask matrix of size(I) with zeros at missing pixels. Now execute:
1. I_mask = I;
2. I_mask(mask == 0) = 0;
3. I_rec = inPainting(I_mask, mask);
The reconstructed image is single channel a matrix I_rec in [0,1] range of the same size as I.
III. Authors
------------
Filippini Luca Teodoro, Department of Computer Science, ETH Zurich, 8092 Zurich, Switzerland. E-mail: fteodoro@student.ethz.ch
Porvaznik Michal, Department of Computer Science, ETH Zurich, 8092 Zurich, Switzer-
land. E-mail: pmichal@student.ethz.ch
Trujic Milos, Department of Computer Science, ETH Zurich, 8092 Zurich, Switzerland. E-mail: mtrujic@student.ethz.ch
近期下载者:
相关文件:
收藏者: