I need simple contrast image similarity, could you tell me what is the recommended algorithm, such as the image below

CodePudding user response:

I don't do the image, but from the perspective of ordinary programmers for

1. The first binarization
2. Angle code, binary point have some distribution trend of unification is adjusted for "light on the turbidity" (statistic + affine transformation)
3. The size code
4. Feature similarity decision (Angle size specifications, then only need to determine characteristics, general algorithm for the cosine similarity)

Ps: the thing you'd better find a professional image to get, we can do only if the application nor the somebody else to do

CodePudding user response:

https://blog.csdn.net/jiangxinyu/article/details/7968139

CodePudding user response:

Point. The point of statute, binarization with some distribution trend of unification is adjusted for "light on the turbidity" (statistic + affine transformation)

Wanted to think, it can use a linear classifier, to get a linear regression line directly, then according to the linear regression line, what do you do after the affine transformation in the feature vector, or directly in the linear regression line vector feature

CodePudding user response:

Throw yolo3

CodePudding user response:

reference 1st floor wanghui0380 response:

I don't do image, but from the perspective of ordinary programmers for

1. The first binarization
2. Angle code, binary point have some distribution trend of unification is adjusted for "light on the turbidity" (statistic + affine transformation)
3. The size code
4. Feature similarity decision (Angle size specifications, then only need to determine characteristics, general algorithm for the cosine similarity)

Ps: the thing you'd better find a professional image to get, we can do only if the application nor the somebody else to do fast

Excuse me, what this code belongs to, I can't search learning materials

CodePudding user response:

To compare with gray histogram

CodePudding user response:

, can't do it like a handwritten Chinese character recognition, no software can identify the, this is similarity algorithm, mathematics haven't break through the boundaries of human beings