euclidean distance calculation in for image forgery detection

i am working on a project and it involves following steps.

 1. RGB to gray conversion of image. 
2. Applying SWT to gray image. 
3. Dividing LL1 subband into overlapping blocks and applying SVD to each block and forming feature matrix. 
4. Lexicographical sorting of this feature matrix. 
5. to find the Euclidean distance of not all the block pairs, but only the very similar blocks that are close to each other (as specified by offset threshold), in the lexicographically sorted feature matrix. condition for offset threshold: abs [index (u) − index (v)] ≤ offset threshold.

I am stuck with this 5th step. Any help in this regard will be highly appreciated.

Answer 1

[KSSV]

doc knnsearch and rangesearch