Mini's Programming: September 2015

It's comparing image similarity using feature matching.
I've used OpenCV and converted c language to java.

The source is as follows:

package kr.co.acronym;

import org.opencv.core.Core;
import org.opencv.core.CvType;
import org.opencv.core.DMatch;
import org.opencv.core.Mat;
import org.opencv.core.MatOfDMatch;
import org.opencv.core.MatOfKeyPoint;
import org.opencv.features2d.DescriptorExtractor;
import org.opencv.features2d.DescriptorMatcher;
import org.opencv.features2d.FeatureDetector;
import org.opencv.imgcodecs.Imgcodecs;

public class Mini {

 public static void main(String[] args) {
  // Set image path
  String filename1 = "C:\\images\\iphone1.jpeg";
  String filename2 = "C:\\images\\iphone2.jpeg";

  int ret;
  ret = compareFeature(filename1, filename2);
  
  if (ret > 0) {
   System.out.println("Two images are same.");
  } else {
   System.out.println("Two images are different.");
  }
 }

 /**
  * Compare that two images is similar using feature mapping  
  * @author minikim
  * @param filename1 - the first image
  * @param filename2 - the second image
  * @return integer - count that has the similarity within images 
  */
 public static int compareFeature(String filename1, String filename2) {
  int retVal = 0;
  long startTime = System.currentTimeMillis();
  
  System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
  
  // Load images to compare
  Mat img1 = Imgcodecs.imread(filename1, Imgcodecs.CV_LOAD_IMAGE_COLOR);
  Mat img2 = Imgcodecs.imread(filename2, Imgcodecs.CV_LOAD_IMAGE_COLOR);

  // Declare key point of images
  MatOfKeyPoint keypoints1 = new MatOfKeyPoint();
  MatOfKeyPoint keypoints2 = new MatOfKeyPoint();
  Mat descriptors1 = new Mat();
  Mat descriptors2 = new Mat();

  // Definition of ORB key point detector and descriptor extractors
  FeatureDetector detector = FeatureDetector.create(FeatureDetector.ORB); 
  DescriptorExtractor extractor = DescriptorExtractor.create(DescriptorExtractor.ORB);

  // Detect key points
  detector.detect(img1, keypoints1);
  detector.detect(img2, keypoints2);
  
  // Extract descriptors
  extractor.compute(img1, keypoints1, descriptors1);
  extractor.compute(img2, keypoints2, descriptors2);

  // Definition of descriptor matcher
  DescriptorMatcher matcher = DescriptorMatcher.create(DescriptorMatcher.BRUTEFORCE_HAMMING);

  // Match points of two images
  MatOfDMatch matches = new MatOfDMatch();
//  System.out.println("Type of Image1= " + descriptors1.type() + ", Type of Image2= " + descriptors2.type());
//  System.out.println("Cols of Image1= " + descriptors1.cols() + ", Cols of Image2= " + descriptors2.cols());
  
  // Avoid to assertion failed
  // Assertion failed (type == src2.type() && src1.cols == src2.cols && (type == CV_32F || type == CV_8U)
  if (descriptors2.cols() == descriptors1.cols()) {
   matcher.match(descriptors1, descriptors2 ,matches);
 
   // Check matches of key points
   DMatch[] match = matches.toArray();
   double max_dist = 0; double min_dist = 100;
   
   for (int i = 0; i < descriptors1.rows(); i++) { 
    double dist = match[i].distance;
       if( dist < min_dist ) min_dist = dist;
       if( dist > max_dist ) max_dist = dist;
   }
   System.out.println("max_dist=" + max_dist + ", min_dist=" + min_dist);
   
      // Extract good images (distances are under 10)
   for (int i = 0; i < descriptors1.rows(); i++) {
    if (match[i].distance <= 10) {
     retVal++;
    }
   }
   System.out.println("matching count=" + retVal);
  }
  
  long estimatedTime = System.currentTimeMillis() - startTime;
  System.out.println("estimatedTime=" + estimatedTime + "ms");
  
  return retVal;
 }
}

I created a new function, compareFeature that is compared two images.
It calculates the distance of the feature in those images.

If you check the last of the source, I decide the similarity by the minimum distance of two images.
The distance value that I choose is 10.
But if you try more and more, the value is more precise.

The return value is the count of the distance under 10.
If it is one more, I think that the two images are similar.

I tested next 3 images.