No specific task , It's just that similar tasks have been published in the image processing group before , I thought it was too difficult at that time , But things are different , After a lot of things, most of my free time is now , I tried to write .
requirement analysis ： Mahjong recognition is also divided into many scenes , I see a single mahjong on the Internet to identify , There are also those who recognize upright mahjong （ Small camera cooperation , Yes ）. These two kinds of recognition are not difficult , Because it bypasses two troublesome steps in image processing , Preprocessing and image segmentation . The final result is that the external verification code recognition is not enough . So the goal I set for myself is to assume that there is a vertical overhead camera right above the mahjong table , Capture and identify cards that have been played .
It is very easy to separate the cards on the mahjong table under the illusion , But there are some problems in practice , For example, the bright spot and shadow of the card caused by the light source , The judgment of the area between the adhesive cards . Then the recognition part is also expected to use feature points matching （SURF,SIFT,FAST Etc ） The way to do it , But it's also a big problem , The matching degree is not high , It's time-consuming and moving . Finally decided to use the symbol on mahjong card , And then combined into a card seal for recognition .
First, the final image of segmentation is given ：
I haven't studied the identification part , But preliminary shape matching was tried , The effect is average , It is intended for later use svm Machine learning and training .
Let's record it today , In the later stage, the problems and solutions encountered in the development process are added .
New blog , The purpose is communication and cooperation , oneself QQ：273651820.