MATLAB Image Processing: Meaning of false rejection rate, false recognition rate, calculation and other error rate Matlab source code (Network First)

The error rejection rate and false recognition rate are two concepts in image processing.

The false rejection rate refers to the error rejection, which refers to intra-class matching. If there are 10 samples of volunteers, each of them has 20 samples. Then, compared with in-class tests, for example, the 20 images in the same category of volunteers on the first day can be matched with each other. Assuming that the matching at is not repeated, the two images can be performed (20*19) /2 times. If 10 volunteers perform this test, 10*(20*19)/2 times. This is the total number of intra-class matches. Pre-set the threshold value to th. If the value matchesTh> TH will be rejected by mistake.

 

The error recognition rate refers to the probability of error acceptance, relative to inter-class matching. Matching between different classes, if the threshold valueTh is less than the preset threshold th, It will be considered to belong to the same class, which is accepted by mistake.

The calculation formula is as follows:

 

You can also refer to the 57th and 20th pages in the document http://ishare.iask.sina.com.cn/f/37004900.html.

Because FRR and far conflict with each other, the probability of equal error rate is equal when the two are equal. When the threshold is th, the corresponding FRR has a value and far has a value. In this way, a group of corresponding FRR and far are obtained. The far is used as the horizontal coordinate, and the FRR is used as the vertical coordinate to draw the curve, that is, the ROC curve. Then draw the Y = x curve. When the two curves intersect, the corresponding horizontal or vertical coordinates are equal error rates. At this time, the corresponding threshold value is optimal.

Typical ROC curve:

No such error rate and ROC Matlab source code have been calculated on the Internet so far. I wrote one for your reference only:

The core code is as follows:

Function out = yaneer (leinei, leijian) ngra = length (leinei); % the number of tests in the class corresponds to the false data rate frrnira = length (leijian ); % The error rate corresponding to the number of inter-class tests, farfrr = []; far = []; Th = 0.1: 0.02: 0.55; % threshold, total XXX for I = 1: length (th) FRR = sum (leinei> TH (I)/ngra; % if it is greater than the preset threshold, FRR = [FRR] is incorrectly rejected; far = sum (leijian <TH (I)/Nira; % far = [far]; End

Here, TH is used to control the pre-set threshold, which can be set as needed. Generally, it is related to the minimum and maximum values of the parameter you enter, or it can be associated with the two most values. In addition, when calculating FRR, You can multiply by 100. Then, the coordinate system is (% ). You can modify it flexibly.

Source code download: http://www.pudn.com/downloads542/sourcecode/graph/detail2238646.html

----------------------- This article is original. For more information, see yanzi1225627.