4. RESTORATION OF IMAGE
Types
Inverse Filtering
Wiener Filtering
Kalman Filtering
Algebraic Approach
Apriori
Blurring function
Noise statistics
5. IMPULSE NOISE EMBEDDED IMAGE (1/2)
Restoration from impulse noise embedded image
Step 1: If the target pixel is noisy, go to step2. Else, go to the next
pixel
Step 2: Replace the noise pixel with a new value.
Local Window
Size: (2M + 1) x (2M + 1)
How to detect noise?
Difference between pixel values from the median of the image
6. IMPULSE NOISE EMBEDDED IMAGE (2/2)
This method will not work fine when the image is too much noisy
The choice of local window may not reflect the global image.
Choice of small local window doesn’t even consider the local regional detail
Wang and Zhang
Two windows of the same size
Around two pixels
One is the target pixel which is noisy
Another is a non-noisy pixel
The non-noisy pixel is selected from a larger sets of candidate
7. MATHEMATICAL DESCRIPTION DEBLUR IMAGE
Shift-Invariant Model - every point in the original image spreads out the same way in forming
the blurry image
Using the convolution Model –
f(x,y) = h(x,y)*g(x,y) + n(x,y)
Here,
g(x,y) = original image
f(x,y) = blurred image
h(x,y) = point spread function or blur function
n(x,y) = noise model
9. INVERSE FILTERING
Fast Fourier Transform and Inverse Fourier Transform give us the solution
Equation of Blur Image,
f(x,y) = h(x,y)*g(x,y) + n(x,y)
Using the Fourier Transformation- convolution can be written in
multiplying the Fourier domain of the point spread function and
original image
F(m,n) = H(m,n) × G(m,n)
G(m,n) = F(m,n) / H(m,n)
g(x,y) = Inverse Fourier (G(m,n))