I implemented the algorithm for local contrast(fast clahe in fpga ). I need to confirm that it improves the image. I have tried to use different analysis methods, but it seems wrong : take original image and original with noise and after compare them. Also tried to use the method of "reference spectrum", but get incorrect results. Could be we use gradients and detected borders, using the methods of the masks of the Sobel or Roberts or Prewitt or((Histogram of Oriented Gradients, SURF) edge or blobs)? Further the more find edges or blobs the more better improved the image?
Thank you very much Aparna and Uche, Vikas thank you too but it is not mathemathic method
First define what you mean by "improves the image". Is it fidelity to the original image? Smoother edges? Increase in SNR? Accurate presentation of grey levels or colours? Correct segment size? Increased continuity of detected edges? etc. Are these "improvements" to be locally defined or global?
Once you have done that, the measurement method will be obvious.
Also, if you want to improve the images so that a human can see them more clearly, the only right answer above is to use perceptual studies. Only if the image is to be used by later-stage algorithms and not by a human is a mathematical answer the right one.
- Badges
- Science topic
- Similar topics
- Control Systems Engineering
- Filtering