Kernel estimation 6. Use the formula of 1D Gaussian function 1 G(x): = 2 e 202...

    

Transcribed Image Text:

Kernel estimation 6. Use the formula of 1D Gaussian function 1 G(x): = 2 e 202 2 to find coefficients of a kernel of size 7 when o=1.4. Hint: x is evaluated in interval [-3-2-10123] [10%] 7. The size of a gaussian kernel is usually chosen to have values in the order of 2 or 3 sigmas, since after that the values of the function are almost zero. In the extreme parts of this kernel (when x is either -3 or 3) how many sigmas it corresponds to? Is the chosen size of 7 a good value? [10%] 8. Approximate the obtained kernel as a fraction of integer numbers. Hint: use 64 as the denominator. [10%] 9. Compute a 7x7 Gaussian kernel using the 1D estimated kernel you estimated in the previous exercise. Remember, this is a separable filter and can be obtained using matrix multiplication. [10%] G = K7x1 * K1x7 10. In class we build a sharpen filter as the sum of original filter + detail. The detail part was built with the original function and a box filter. Create a new kernel for sharpening but this time uses a gaussian filter. [10%] Kernel estimation 6. Use the formula of 1D Gaussian function 1 G(x): = 2 e 202 2 to find coefficients of a kernel of size 7 when o=1.4. Hint: x is evaluated in interval [-3-2-10123] [10%] 7. The size of a gaussian kernel is usually chosen to have values in the order of 2 or 3 sigmas, since after that the values of the function are almost zero. In the extreme parts of this kernel (when x is either -3 or 3) how many sigmas it corresponds to? Is the chosen size of 7 a good value? [10%] 8. Approximate the obtained kernel as a fraction of integer numbers. Hint: use 64 as the denominator. [10%] 9. Compute a 7x7 Gaussian kernel using the 1D estimated kernel you estimated in the previous exercise. Remember, this is a separable filter and can be obtained using matrix multiplication. [10%] G = K7x1 * K1x7 10. In class we build a sharpen filter as the sum of original filter + detail. The detail part was built with the original function and a box filter. Create a new kernel for sharpening but this time uses a gaussian filter. [10%]