Face detection is to judge whether there is a plane image in dynamic scene and complex background and separate the plane image. Generally, there are several methods:
(1) reference template method
Firstly, one or several templates of standard faces are designed, and then the matching degree between the samples collected by the test and the standard templates is calculated, and the threshold value is used to judge whether there is a face or not.
(2) face rule method
Because the face has certain structural distribution features, the so-called face rule method is to extract these features and generate corresponding rules to judge whether the test sample contains a face or not.
(3) sample learning method
In this method, the artificial neural network is used in pattern recognition, that is, the classifier is generated by learning the sample sets of opposite image and non-plane image.
4. Skin color model method
This method is based on the appearance of skin color distribution in the color space is relatively concentrated to detect the law.
5. Feature child face method
In this method, all face sets are regarded as a face subspace, and the existence of face images is judged based on the distance between the detection sample and its projection in the subspace.
It is worth mentioning that the above five methods can also be used comprehensively in the actual detection system.
(2) face tracking
Face tracking refers to the dynamic target tracking of the detected face. The method based on model or the method based on the combination of motion and model is adopted. In addition, the use of skin color model tracking is also a simple and effective means.
(3) face comparison
Face comparison is to confirm the identity of the detected face image or search the object in the face image library. This means, in effect, comparing the sampled plane images with those in stock one by one to find the best match. Therefore, the description of plane image determines the specific method and performance of plane image recognition. Feature vectors and face pattern templates are mainly used for description:
(1) eigenvector method
In this method, the size, position, distance and other attributes of the contour of five facial features of iris, nose wing and corner of mouth are determined first, and then their geometric feature quantities are calculated, which form an feature vector to describe the facial image.
(2) surface pattern template method
This method stores several standard plane image templates or plane image organ templates in the library, and matches all pixels of the sampled plane image with all templates in the library by normalized correlation measure. In addition, there is a method to combine pattern recognition autocorrelation network or feature with template.
The core of face recognition technology is "local human feature analysis" and "pattern/neural recognition algorithm". This algorithm is a method that utilizes various organs and characteristic parts of human face. For example, multiple data of corresponding geometric relations form identification parameters and compare, judge and confirm all original parameters in the database. The judgment time is generally required to be less than 1 second.