Recently, due to the needs of large-scale homework, opencv was used for face recognition.
Generally speaking, recognition is divided into two parts, namely Face Detection -> Face Recognition. Opencv provides a face detection sample, and has a relatively mature face training file. This is a very mature face recognition tool I know, and it has been used in some cutting-edge 3D love action games. Here, we mainly make certain modifications to the sample provided by it, and save the face as a picture for the next step of processing.
The processing done here: detect the face -》resize is 100x100 -》histogram equalization -》takes the 70x70 area (removes hair and other parts). I encountered a problem here: I used the same code, and under debug, I did not missed face detection, but using opencv2.0 will cause more false detections, while using opencv2.1 to run the same code, which has much less false detection, which has the same effect as those in the sample directory. This has not been studied in depth, and my classmates also encountered the same problem.
The next thing is to extract the principal components. If you use the PCA algorithm, there are two main ideas:
1. Leave everyone's faces to PCA to extract the principal components, and then perform principal components analysis for each new picture to reduce the dimensions, and judge the difference between the two faces by Euclidean distance.
2. This time, a person's face is handed over to PCA to extract the principal components. Then, for each new picture, the principal component is analyzed first to reduce the dimension, and then map back to the original picture in reverse, and compare the reconstructed picture with the original picture. If the difference is small, it means it is a face, and if the difference is large, it means it is a different face.
I used the first method in the program. The analyzed main components are called characteristic faces. Each face can be compared after dimensionality reduction. Here I manually perform the simplest classification, calculate the centers of several faces, and then compare other faces with them. If it is greater than one threshold, it is considered to be different faces.
Before comparing the new pictures, you must also check, identify and give the results, and finally you can correctly detect your face in a group photo!
Attach reference materials used when doing homework:
/?tid=32283###