I'm trying to implement retina filter implementation that operates on the principle of human vision. Therefore they are several filters which recognize contours and movements in image sequences.
One of them is called OPL filter and has the transfer function in the paper attached:
Is there any possibility to find an appropriate filter mask of filter kernal that represent this transfer function in the paper attached.and please also how I may apply this transfer function, i.e, on one image or image sequence.
and also how can i use temporal derivation. Thanks
Please check the paper attached.
Hello,
though slightly obscured the [-z(exp-1) 2 -z] is describing a 3 x 1 "matrix" filter - something you normally use in a 3x3 or 5x5 configuration for edge detection, low/high pass filtering etc. Not exactly a standard algorithm, but understandable. You might even be able to improve results by using a 3x3 filter with appropriate coefficients but - regarding the complete formula in the reference -you would have to adapt the "surrounding" formula.
To apply such a filter you basically have to "fold" this filter with the image - nothing more than sliding the matrix over the picture - pixel by pixel. Every tool capable of "folding" such a matrix should be able to perform the basic operation.
The second step to implement the OPL would be to implement the "rest" of he equation for each pixel - using the result from step 1 to replace the 3 [...] terms. Should not pose a too complicated task but will most likely not be available as an out-of-the-box solution :)
Looks like fun.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Human-Computer Interaction