Picture is such, a total of four of the key direction of the picture inside of the nine key more or less have some different, that is a different size, I use the method of template matching (matchTemplate) to a fixed template figure to match though can match the key just slightly change to identify the direction is wrong,,,,,,,,

In addition to the template matching method to still have what good method can let me identify the direction of these keys??????

Using the opencv 3.2

CodePudding user response:

This seems very simple, horizontal vertical count classification can be realized, need not the too complicated

CodePudding user response:

The match should be not so bad as you say, don't know whether your judgment logic has a problem? When there is a match degree

CodePudding user response:

Not necessarily fixed 9 key, sometimes is 3 4 keys such as this


This is my judgment logic

This is a graph template
I opencv novice hope no other comment

CodePudding user response:

refer to the second floor robertbo response:

match should be not so bad as you say, don't know whether your judgment logic has a problem? When there is a match degree

reference 1/f, ArWen response:

this seems very simple, horizontal vertical counting classification can be realized, need not the too complicated

I posted the template above figure out don't know the steps out of the question

CodePudding user response:

Template matching with binary images, you have to cycle again find matching degree is the highest, is not to find the first must be the most appropriate, the processing logic inside need to have a good reason

CodePudding user response:

Has been solved, the interpreter is right to say that above, and there is something wrong with the processing logic leads to the recognition rate is low

CodePudding user response:

refer to 6th floor qqabxiaojing response:

has been solved, the interpreter said above is correct,,, there is something wrong with the processing logic leads to the recognition rate of lower

Can post to have a look

CodePudding user response:

Can try neural network algorithm