Ask Your Question
2

Image matching problem

asked 2012-11-24 08:16:35 -0500

Szippy gravatar image

updated 2012-11-25 05:41:26 -0500

Hello

I want to amtch images of banknotes of witch i have photos and compare them to the image taken recently, so that I can determin witch kinde of banknote I took a photo of.

I am loading the banknotes into a vector<mat> and iterate thru them and compare each to the photo just taken. I have used the FLANN based amtching example as a starting point, but the matching done there has a lot of errors points are matched utterly wrong, and that messes up my matching process. The images are kinda similar, as in the banknotes have some similarities, but that shoudn't be the problem, as the example shows the points are matched exactly. So my question would be is there an optimal way to match banknotes? Maybe there is a flaw in my matching settings.

These are some of my sample Images:

image description image description image description image description

And these are the two images that I want to compare them to: image description image description

with this code i load all my files:

for(std::set<string>::iterator Name = ListOfFileNames.begin() ; Name != ListOfFileNames.end() ; ++Name)
{

    string path = "ron/";
    string name = *Name;
    if(name.length()>4)
    {
        path.append(name);
        IplImage *img=cvLoadImage(path.c_str());
        cv::Mat tempImage(img,CV_LOAD_IMAGE_GRAYSCALE);
        cvtColor(tempImage, tempImage, CV_RGB2GRAY);
        loadedImages.push_back(tempImage);
        loadedImagesNames.push_back(name);
    }
}

And with the Orb matcher I compare the new files to the dataset:

int compareOrb(cv::Mat *img_1t, cv::Mat *img_2t){

int minHessian = 800;
Mat img_1 = *img_1t;
Mat img_2 = *img_2t;

OrbFeatureDetector detector( minHessian );

std::vector<KeyPoint> keypoints_1, keypoints_2;

detector.detect( img_1, keypoints_1 );
detector.detect( img_2, keypoints_2 );

//-- Step 2: Calculate descriptors (feature vectors)
OrbDescriptorExtractor extractor;

Mat descriptors_1, descriptors_2;

extractor.compute( img_1, keypoints_1, descriptors_1 );
extractor.compute( img_2, keypoints_2, descriptors_2 );

//-- Step 3: Matching descriptor vectors using FLANN matcher
if(descriptors_1.type()!=CV_32F) {
    descriptors_1.convertTo(descriptors_1, CV_32F);
}

if(descriptors_2.type()!=CV_32F) {
    descriptors_2.convertTo(descriptors_2, CV_32F);
}
FlannBasedMatcher matcher;
std::vector< DMatch > matches;
matcher.match( descriptors_1, descriptors_2, matches );

double max_dist = 0; double min_dist = 100;

//-- Quick calculation of max and min distances between keypoints
for( int i = 0; i < descriptors_1.rows; i++ )
{ double dist = matches[i].distance;
if( dist < min_dist ) min_dist = dist;
if( dist > max_dist ) max_dist = dist;
}
/*
printf("-- Max dist : %f \n", max_dist );
printf("-- Min dist : %f \n", min_dist );
*/
//-- Draw only "good" matches (i.e. whose distance is less than 2*min_dist )
//-- PS.- radiusMatch can also be used here.
std::vector< DMatch > good_matches;

for( int i = 0; i < descriptors_1.rows; i++ )
{ if( matches[i].distance < 2*min_dist )
{ good_matches.push_back( matches[i]); }
}

//-- Draw only "good" matches
Mat img_matches;
drawMatches( img_1, keypoints_1, img_2, keypoints_2,
    good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
    vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );

//-- Show detected matches
imshow( "Good Matches", img_matches );

for( int i = 0; i < good_matches.size(); i++ )
{ printf( "-- Good Match [%d] Keypoint 1: %d  -- Keypoint 2: %d  \n", i, good_matches[i].queryIdx, good_matches[i].trainIdx ); }

waitKey(0);

return good_matches.size();

}

And with this loop I check them all:

    int best = 0;
int bestId = -1;
int t;
for(int i=0;i<loadedImages.size();i++)
{
    t =compareOrb(&loadedImages.at(i),&image5);
    //t=compareSurf ...
(more)
edit retag flag offensive close merge delete

Comments

I don't know the time constraint of your application, but for example you can switch (temporaly) to the BruteForce method and impose the crossCheck flag, so that you always have two features that match in a reciprocal way. This may be helpful to see if there are some problem in your code. Also if u can share part of the code where u do the matching it would be helpful to see if everything is done correctly. Here is the doc for the BruteForce approach http://docs.opencv.org/modules/features2d/doc/common_interfaces_of_descriptor_matchers.html#bfmatcher-bfmatcher

Simone gravatar imageSimone ( 2012-11-25 04:02:35 -0500 )edit

In order to give you good advice we will need to see some images you are trying to match. Hundreds of matching approaching exist, but approach that is good for one problem may be completely inappropriate for another.

Michael Burdinov gravatar imageMichael Burdinov ( 2012-11-25 04:57:38 -0500 )edit

I have added the images and the code.

Szippy gravatar imageSzippy ( 2012-11-25 05:41:57 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2012-11-26 05:07:57 -0500

Michael Burdinov gravatar image

updated 2012-11-26 05:33:55 -0500

Transformation you are trying to find is a very complicate one. Even if banknotes were perfectly flat it seems like (bounded) perspective transformation. Folding those banknotes means that the match should be elastic. The fact that transformation is bounded may help, but still you are trying to solve very difficult problem.

If your task is to make a choise between small number of very distinctive objects, I would recommend you to try other approaches. For example try to find colors that belong to only one of the banknotes. Or try to find some digits. Finding single digit is much much simpler than matching whole image.

edit flag offensive delete link more

Comments

Digit matching is no an option as it is banknotes i want to check and if I have some numbers drawn on a piece of paper and my software tells me it's money that is a bad result

Szippy gravatar imageSzippy ( 2012-11-29 23:54:24 -0500 )edit

I expect this would not be an issue, as the OCR would need to be trained to the particular font and would under no circumstances pick up common handwriting.

Christopher gravatar imageChristopher ( 2014-10-19 11:16:30 -0500 )edit
Login/Signup to Answer

Question Tools

Stats

Asked: 2012-11-24 08:16:35 -0500

Seen: 849 times

Last updated: Nov 26 '12