Unsharp Masking
- Process flow:
It is simply the input image Laplacian filter (high pass filter) to obtain a high-frequency signal, and to some extent to the original signal obtained by adding the corrected result. Is the result of high-frequency signal thus processed original image is highlighted.
The image frequency signal is typically a Gaussian filter (low pass filter) may be obtained by subtracting the original image. The following is a sample processing:
It can be seen that the edge (high frequency signal) is enhanced in FIG highlighted.
-
Formula expression:
Which x(n, m)
means that the input image, i.e. the image to be corrected; y(n, m)
refers to an output image, i.e. corrected image; z(n, m)
refers to a correction signal; λ
is a positive number of the scaling factor.
z(n, m)
You can be obtained by the following formula:
Adaptive Unsharp Masking
- Process flow:
On the lower half of FIG flow can be seen to provide a two bilateral filtering operation, respectively horizontal and vertical directions:
Corresponding output y(n, m)
can be expressed as:
Where λx(n, m)
and λy(n, m)
are respectively corresponding to the scaling factor signal.
Goal is to use an adaptive algorithm recursively update these parameters, can enhance the image region smooth little or no enhancement, the contrast medium to enhance the maximum area, and large area contrast is enhanced only moderately.
The following represented by a vector scaling factor and the correlation signal:
Corresponding output can be expressed as:
- Cost Function:
For ease of implementation and analysis traceability of adaptive filters, we define a method of measuring local dynamic image (local dynamics) using a simple linear high pass filter g()
output, supports 3×3
pixel, as shown in FIG.
Choosing linear operator g()
rather than the other metrics, such as the local variance, because the operator using the adaptive algorithm is relatively simple to produce.
Whole metric equation is as follows:
Algorithm using Gauss-Newton adaptive algorithm to reduce the local dynamic and desired by g()
the error between the local dynamic obtained.
First, based on the activity level of the image measured (activity level), each of the pixels of the input image divided into three categories a, the activity level is calculated on a given block of 3 × 3 pixel local variance :
After one refers to the average luminance level.
Based on this basic principle, define the required level of activity in the output image is:
Finally, the cost function is:
Reproduced in: https: //www.jianshu.com/p/604eae9d78be