Revision history [back]

You can prune your results, in multiple ways, some of them are:

• Cross-check: Make a cross-check of your matches (if you use BruteForce-matcher you can set this option)
• Ratio-Test of your matcher: See e.g. http://answers.opencv.org/question/4829/how-to-filter-freakbruteforcematcher-result
• Trickier: Compute transformation (homography) matrix, warp the image according to it and see how many overlap you get, either by computing directly the root-means-square-error or by computing the cross-correlation (matchTemplate()) from your object and the warped image and see if you get a high peak or not.

You can prune your results, in multiple ways, some of them are:

• Cross-check: Make a cross-check of your matches (if you use BruteForce-matcher you can set this option)
• Ratio-Test of your matcher: See e.g. http://answers.opencv.org/question/4829/how-to-filter-freakbruteforcematcher-result
• Trickier: Compute transformation (homography) matrix, warp the image according to it and see how many overlap you get, either by computing directly the root-means-square-error or by computing the cross-correlation (matchTemplate()) from your object and the warped image and see if you get a high peak or not.
• Use a completly different learning-based approach by e.g. using multiple CascadeClassifiers or Bag-of-Words with a classifier (http://answers.opencv.org/question/8677/image-comparison-with-a-database/#8686)