cv::resize and cv::warpAffine produce different results (Bugfix #4133)

Added by Yida Wang about 10 years ago. Updated almost 10 years ago.

Status:Cancelled Start date:2013-08-12
Priority:High Due date:
Assignee:Roman Donchenko % Done:

100%
Category:imgproc, video
Target version:-
Affected version:branch '2.4' (2.4-dev) Operating System:Any
Difficulty:Hard HW Platform:Any
Pull request:

Description

Code to reproduce:

 1Mat tst(1, 2, CV_8U);
 2tst.at<uchar>(0,0) = 0;
 3tst.at<uchar>(0,1) = 128;
 4
 5Mat transf(2,3, CV_32F);
 6transf.at<float>(0, 0) = 5;
 7transf.at<float>(0, 1) = 0;
 8transf.at<float>(0, 2) = 0;
 9transf.at<float>(1, 0) = 0;
10transf.at<float>(1, 1) = 1;
11transf.at<float>(1, 2) = 0;
12
13Mat r1,r2;
14resize(tst, r1, Size(), 5, 1, INTER_LINEAR);
15warpAffine(tst, r2, transf, r1.size(), INTER_LINEAR, BORDER_REPLICATE);
16
17cout << "r1: " << r1 << endl;
18cout << "r2: " << r2 << endl;

Ouput:

r1: [0, 0, 0, 26, 51, 77, 102, 128, 128, 128]
r2: [0, 24, 52, 76, 104, 128, 128, 128, 128, 128]

The problem lies in the affine Metrix, it is wrongly designed, to get a fine affine Metrix, you need to set up three points on the source and destination images to compute an affine transformation using:

transf = getAffineTransform( srcTri, dstTri );

and the source code should be rewrite as:
transf.at<float>(1, 1) = 0;
transf.at<float>(1, 1) = 1;

So,
New Ouput:

r1: [0, 0, 0, 26, 51, 77, 102, 128, 128, 128]
r2: [0, 0, 0, 24, 52, 76, 104, 128, 128, 128]

Associated revisions

Revision b98a54fb
Added by Maksim Shabunin over 9 years ago

Merge pull request #4133 from ruslo:interface.sample

History

#1
Updated by Aly Osama about 10 years ago

Could you run this testcase and tell me your result

Mat tst(2, 2, CV_8U);
tst.at<uchar>(0,0) = 1;
tst.at<uchar>(0,1) = 4;
tst.at<uchar>(1,0) = 4;
tst.at<uchar>(1,1) = 7;

Mat transf(2,3, CV_32F);
transf.at<float>(0, 0) = 2;
transf.at<float>(0, 1) = 0;
transf.at<float>(0, 2) = 0;
transf.at<float>(1, 0) = 0;
transf.at<float>(1, 1) = 2;
transf.at<float>(1, 2) = 0;

#2
Updated by Yida Wang about 10 years ago

r2: [1, 3, 4, 4, 4, 4, 4, 4, 4, 4;
3, 4, 6, 6, 6, 6, 6, 6, 6, 6]

#3
Updated by Aly Osama about 10 years ago

it should be as r1 so the problem is at wrapAffine function

r1:
[ 1, 2, 3, 4;
2, 2, 4, 5;
3, 4, 5, 6;
4, 5, 6, 7]

#4
Updated by Ankit Dhiman about 10 years ago

I tried using warpPerspective function.
Mat tst(2, 2, CV_8U);
tst.at<uchar>(0,0) = 1;
tst.at<uchar>(0,1) = 4;
tst.at<uchar>(1,0) = 4;
tst.at<uchar>(1,1) = 7;
Mat transf1(3,3, CV_32F);
transf1.at<float>(0, 0) = 2;
transf1.at<float>(0, 1) = 0;
transf1.at<float>(0, 2) = 0;
transf1.at<float>(1, 0) = 0;
transf1.at<float>(1, 1) = 2;
transf1.at<float>(1, 2) = 0;
transf1.at<float>(2, 0) = 0;
transf1.at<float>(2, 1) = 0;
transf1.at<float>(2, 2) = 1;
Mat r1,r3;
resize(tst, r1, Size(), 2, 2, INTER_LINEAR);
warpPerspective(tst,r3,transf1,r1.size(), INTER_LINEAR,BORDER_REPLICATE);

Output : [1 3 4 4 ;
3 4 6 6;
4 6 7 7;
4 6 7 7]
If I perform silmilar steps in Matlab. Output by imresize and imtransform are same. Has this issue been resolved?

#5
Updated by Vadim Pisarevsky almost 10 years ago

not sure about the essence of the bug report

  • Status changed from Open to Cancelled

Also available in: Atom PDF