Chapter 5

Question 1

What is neighbourhood processing?

Answer 1

To calculate an output pixel value, the input pixel and it’s neighbours are used.

Question 2

Name important neighbourhood rank filters:

Answer 2

1) the median filter
2) the minimum filter
3) the maximum filter
4) difference between min and max

Question 3

What is correlation?

Answer 3

A type of filter (in this case called a kernel) that is applied to every pixel, and the kernel has weights.

Question 4

What is a gaussian kernel?

Answer 4

A kernel with weights derived from gaussian distribution - most weight on the pixels close to the center.

Question 5

What is template matching?

Answer 5

An application of correlation. The kernel defines an object we are looking for - called a template. Each pixel in the new picture has a value that corresponds to how well the original picture mathes the template. One problem is that it tends to create bias for bright areas.

Question 6

How is template matching bright spot problems circumvented?

Answer 6

By normalization. Think of the template and image patch as represented by vectors in n (number of pixels in template) dimensional space. This can be normalized (see notes)

Question 7

What is the normalized cross correlation?

Answer 7

NCC = correlation / ( length template * length image patch)

Question 8

What is a picture gradient?

Answer 8

The change in pixel values between pixels - one in the x-x direction and one in the x-y direction. These together spand the gradient plane.

Question 9

How to calculate picture gradients:

Answer 9

By applying filters:

1) prewitt (-1,-1,-1,0,0,0,1,1,1) either horizontal or vertical.
2) sobel (-1,-2,-1,0,0,0,1,2,1)