Not getting correct results for copy move forgery detection...

I am working on copy-move forgery detection using feature vector using radix sort. 1. I have divided the image in overlapping blocks:) 2. the corners of each block is stored:) 3. feature vector for each block is found:) 4. these features are sorted using radix sort:) 5. the similar regions will give us same feature vectors too...so when sorted, similar feature vectors are brought together:) 6. the difference of the positions of every of every pair of adjacent feature vectors in the sorted list is computed:) 7. the accumulated number of each of the shift vectors is found:) now I am facing problem:( based on large accumulated number of shift vectors,how should I detect the corresponding matched blocks? how should I decide the threshold?