Quote:
Originally Posted by asmit
Can anyone help me with the code on "image restoration using neural networks"...?......preferabbly VC++ platform...
Initially the project is to be degraded (blurred) by adding noise and then it is to be restored (preferabbly using neural networks )..
|
Blurring tends to "spread" information in the image. There are a number of standard deblurring techniques which often work quite well. Generally, they work by re-arranging the blurred information in a region around the pixel of interest.
To do this with neural networks (or any other machine learning technique), I suggest constructing a neural network to operate on a small region (say, 7x7 pixels) at a time, "fixing" the pixel in the center by predicting a new value for it. Training samples would be composed of pixels in the surrounding neighborhood as inputs, and the center pixel as the target. Inputs would come from a blurred image, while the target would be drawn from the original, unblurred image.
To deblur a new image, the neural network would scan over the blurred image, one pixel at a time, predicting new values for each pixel, which are used to construct a new, sharpened image.
-Will Dwinnell
Data Mining in MATLAB
image restoration using neural networks
