"Fuzzy" There are many algorithms, there is something called "Gaussian Blur" (Gaussian Blur). It normally distributed (also known as "Gaussian") for image processing.
The so-called "fuzzy", will be understood as averaging each pixel the peripheral pixels.
Normal distribution
The normal distribution, the closer to the center point, the larger the value, the more away from the center, a shorter interval.
Calculating the average time, we only need to "center point" as the origin, according to its position in the other points on the normal curve, the weight distribution, a weighted average can be obtained. Normal distribution is obviously a desirable weight distribution pattern.
1.3.2 Gaussian function
How to reflect the normal distribution? You need to use a high function to achieve.
Above normal distribution it is one-dimensional, but the image is two-dimensional, so we need a two-dimensional normal distribution.
Gaussian filter is an advanced version of the average filter, the only difference is that each element of convolution kernels mean filter are the same, and the convolution kernel Gaussian filter element Gaussian distribution.
Gaussian filter is based on a two-dimensional Gaussian distribution function, and therefore introduces a two-dimensional Gaussian distribution function. And two-dimensional Gaussian distribution function as follows:
With this function, you can calculate the weight of each point is heavier.
In the image filtering, we also briefly talked about the Gaussian filter:
Image filtering
1.3.3 obtain weight matrix
Assumed center point coordinates are (0,0), the distance coordinate its nearest eight points are as follows:
To calculate the weight matrix, it is necessary to set the value of σ. Assuming σ = 1.5,
The blur radius weight of 1 matrix as follows:
Right nine points equal to the sum of the weight 0.4787147 (nine point addition), only if the calculated weighted average of the nine points, must have their weights equals 1, so the above nine values are also divided by 0.4787147 to obtain a final weight matrix.
Divided by the total value of this process is also called "normalization issues,"
the purpose is to make the total value of the filter weight is equal to 1. Otherwise, the use of the filter will be greater than the total value of a partial image light, the filter is less than 1 will dim image.
Calculate fuzzy values
With the weight matrix, we can calculate the value of a Gaussian blur.
Suppose prior nine pixels, gray value (0-255) as follows:
Each point is multiplied by its own weight value:
get
These values add up to 9, it is the center point of the value of the Gaussian blur.
Repeat this process for all the points, you get the image of Gaussian blur. For color images, it is required to do three RGB channels are Gaussian blur.