KEMBAR78
Convolution and Correlation: Computer Vision CITS4240 | PDF | Convolution | Function (Mathematics)
Scribd
Upload
118 views4 pages

Convolution and Correlation: Computer Vision CITS4240

Convolution and correlation are methods to apply spatial filters to an image. Convolution involves flipping a filter function and sliding it across an image to integrate the overlapping areas. Correlation is similar but does not require flipping the filter function. Both operations are used in computer vision tasks to analyze image features.

Uploaded by

kuzeysc
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
118 views4 pages

Convolution and Correlation: Computer Vision CITS4240

Convolution and correlation are methods to apply spatial filters to an image. Convolution involves flipping a filter function and sliding it across an image to integrate the overlapping areas. Correlation is similar but does not require flipping the filter function. Both operations are used in computer vision tasks to analyze image features.

Uploaded by

kuzeysc
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Computer Vision CITS4240

School of Computer Science & Software Engineering The University of Western Australia

Convolution and Correlation


Convolution and correlation are the methods by which we apply spatial lters to an image.

Convolution
Convolution of two functions f (x) and g(x), usually denoted by f (x) g(x) or f (x) g(x), is given by f (x) g(x) = f ()g(x )d

where, in our context, f (x) is typically the image, g(x) is the lter, and is a dummy variable of integration. In the discrete case the equation above becomes a summation f (x) g(x) = f ()g(x ).

Informally: convolution involves ipping the second function about the origin and then sliding it past the rst function, integrating the product of the overlapping sections of the functions. Figures 1 and 2 give graphical illustrations of the convolution process for the continuous and discrete cases.

Correlation
Correlation of two functions f (x) and g(x), usually denoted by f (x) g(x), is given by

f (x) g(x) =

f ()g(x + )d

where f (x) denotes the complex conjugate of f (x). Taking the complex conjugate simply involves negating the imaginary component. For a real valued function this means no change. Correlation tells you how similar the signal is to the lter at any point. 1

f ()
1

g()

1/2 1 1

g()

g(x )

f ()g(x )
as x is increased g slides across

f ()g(x )

f (x) g(x)

1/2

Figure 1: Graphical illustration of the convolution process. At the top we see the functions f () and g(). The second row shows the ipped version of g, g() and when g is shifted by x, g(x ). The third row shows the inner product between f () and g(x ), which lead to the nal result shown at the bottom.

1 x

0.8

0.7

0.3

r(x) =

p1 (t)p2 (x t)

p1 (t)
2 1 0 1 2

x = 1, p2 (1 t)
2 1 0 1 2

r(1) = 0

x = 0, p2 (0 t)
2 1 0 1 2

r(0) = 0.56

x = 1, p2 (1 t)
2 1 0 1 2

r(1) = 0.24

0.56 Result r(x) 0.24

Figure 2: Graphical illustration of convolution in the discrete domain.

Correlation is similar to convolution except that one does not need to ip g(x) about the origin. For real-valued functions where the second function (the lter) is symmetric convolution and correlation are identical. Where the second function is anti-symmetric the convolution will be the negative of the correlation. Almost all ltering functions used are either symmetric or anti-symmetric. The term convolution is often used quite loosely in computer vision. Very often when people say convolution what they typically mean, or implement, is actually correlation. This usually does not matter because we usually work with real-valued data and with lters that are symmetric or anti-symmetric. However, be aware of this and look out for situations where the distinction is important, especially when using complex valued data.

You might also like