Gaussian Blur is one of the many fuzzy algorithms, the so-called blur, smoothing the image, eliminating the difference between pixels, the most easily thought of is the mean value smoothing.
1. Mean value Blur
The mean Blur is the average of the pixels around the target pixels. For example,
The pixel matrix.
|1|1|1|
|1|2|1|
|1|1|1|,
After the mean value is blurred, it becomes
|1|1|1|
|1|1|1|
|1|1|1|,
This blurs the difference between pixels, but it is obviously flawed because the farther away the pixel has less impact on the target pixel.
Therefore, the farther away from the target pixels, the greater the effect on the target pixels, and the smaller the inverse.
So, this very fits the normal distribution, so we first Sir into a Gaussian nucleus, the formula for
G (x, y) = (1/2*pi*sigma^2) ^n* (e^ (-(x^2+y^2/) 2*sigma*sigma)
where Sigma is the standard deviation, n is demension, and according to the Gaussian kernel function, it is possible to generate kernel,
For example, 5*5 's kernel is
6.5857407E-6 4.247807E-4 0.0017035368 4.247807E-4 6.5857407E-6
4.247807E-4 0.027398387 0.10987825 0.027398387 4.247807E-4
0.0017035368 0.10987825 0.44065478) 0.10987825 0.0017035368
4.247807E-4 0.027398387 0.10987825 0.027398387 4.247807E-4
6.5857407E-6 4.247807E-4 0.0017035368 4.247807E-4 6.5857407E-6
(usually we only need to calculate (6*sigma+1) (6*sigma+1) size of the kernel)
In this way, we have a kernel, we can do convolutions with the original image.
Gaussian Blur in the SIFT