opencv3.0 illuminationChange Issue and bug fix

asked 2014-08-22 23:30:53 -0500

vsooda gravatar image

updated 2014-08-23 00:14:21 -0500


I have find a bug in illuminationChange sample. when I use the illuminationchange sample code to test the function.The result is wrong. image description

I have tried to fix the bug and it looked as if I were succeeding. The result is as follow: image description

The bug is in the file sources\modules\photo\src\seamless_cloning.hpp (line 549 and 553), in the functon Cloning::illum_change. The magnitude of the place outside of mask is [0, 0, 0] and if calucalate: pow(mag, -1 * beta, multy_temp); The value of multy_temp will be "-1.#IND".

what I have do is do filter the mag value as follow:

for (int i = 0; i < mag.cols; i++)
        for (int j = 0; j < mag.rows; j++)
            if (<cv::Vec3f>(j, i)[0] == 0)
      <cv::Vec3f>(j, i)[0] = 1e-8;

            if (<cv::Vec3f>(j, i)[1] == 0)
      <cv::Vec3f>(j, i)[1] = 1e-8;

            if (<cv::Vec3f>(j, i)[2] == 0)
      <cv::Vec3f>(j, i)[2] = 1e-8;

This is my solutionn. There must be some better methods to fix the bug.

edit retag flag offensive close merge delete



could you add a link to the sample ? can't find it.

berak gravatar imageberak ( 2014-08-23 01:17:42 -0500 )edit

And provide a fix if you find one through this topic! Thanks in advance!

StevenPuttemans gravatar imageStevenPuttemans ( 2014-08-23 04:31:57 -0500 )edit

@berak , There is bug in the api illuminationChange" not the sample. The api can find in . The problem is in the Cloning::illum_change functions.

vsooda gravatar imagevsooda ( 2014-08-23 08:41:09 -0500 )edit

@StevenPuttemans . Adding the code in my question after line 544 can fix the bug.

vsooda gravatar imagevsooda ( 2014-08-23 08:44:51 -0500 )edit

@vsooda, i know, just wanted to play around with it ;)

you made it pretty obvious already. wherever mag is 0, pow(mag, -something) is INF. well understood.

berak gravatar imageberak ( 2014-08-23 09:30:27 -0500 )edit

@berak . Thank you. This is my first time to report bug. The test data are in this page . You can simplely read the source image and mask, then call the api illuminationChange. And the show the result. You can also see the test code in (TEST(Photo_SeamlessClone_illuminationChange, regression))

vsooda gravatar imagevsooda ( 2014-08-23 10:07:58 -0500 )edit

@vsooda, thanks a lot, i somehow missed that.

(and no fear, even if you're new to it, all of them quite good questions, imho, keep on with it ;)

berak gravatar imageberak ( 2014-08-23 10:17:53 -0500 )edit

@berak, thank you. I am using opencv for my researsh recently. If I find any other bug, I will report and fix it.

vsooda gravatar imagevsooda ( 2014-08-23 20:28:08 -0500 )edit