Home > Technology > Technical Articles > The technical principle and recognition process detailed analysis of face recognition
The technical principle and recognition process detailed analysis of face recognition
2014-08-21 09:43:53

Face recognition consists of three parts:

(1) face detection

Face detection is to dynamic scenes with complex background in judging the face, and to isolate the face. Generally have the following methods:

The reference template method

First, one or several standard face template, template matching degree between the sample and standard and calculation of the test collection, and through the threshold to determine whether there is a face;

face rule method

Because the human face has some characteristics of the structure distribution, method called face rules that these features are extracted to generate the corresponding rules to determine whether a test sample containing the face;

Sample learning method

This method is using the method of artificial neural network in pattern recognition, which is based on the sample set and non face like image classifier learning sample set;

The skin color model

This method is based on the appearance of skin color in the color space distribution of the relative concentration of the rules to detect.

The eigen face method

This method is all face image set as a face subspace, and based on the detected samples and between the projection holes between the distance to determine whether there is a.

It is worth mentioning that, these 5 methods can be used in practical detection system.

(2) face tracking

Face tracking is to track the detected features of dynamic target. The specific model based method or the method based on the combination of movement and model. In addition, using a simple and effective means is the skin color model tracking.

(3) face alignment

Face alignment is to be detected look like identification or in the face of target searching in the library. This fact is, the sampling to the face and inventory turn face like comparison, and find out the best matching object. So, like the description determines the face and performance of specific identification methods. At present mainly uses the feature vector and face template two description methods:

The eigenvector method

The method is to determine the eyes, nose, mouth and other facial features outline size, position, distance and other attributes, and then calculate the geometric characteristics of them, and these features to form the feature vector describing the surface like a.

The face template method

The method is stored in the database of standard face image template or facial organs template, in comparison, the sample surface like all pixels in the library all the template matching using normalized correlation measure. In addition, there are the self correlation network or features and template combined with pattern recognition.

The core of face recognition technology for analysis of "actual" local body character and "graphics / neural recognition algorithm." This algorithm is a method of using human face organs and parts. If the corresponding geometric relationship of multiple data form the original all the parameters of the identification parameters in the database to compare, judgment and confirmation. The judgment in less than a second general.

The process of face recognition

Generally divided into three steps:

(1) the first build the face facial profile. The face camera acquisition unit personnel face image file or take photos of them forming surface like a file, and the face image file generation faces (Faceprint) coding storage.

(2) the human face image. The camera capture the current staff's face, or take photographs of input, and the current face image file generation mask code.

(3) ratio on the face coding and archival current. Face encoding mask code and archival is the current face of retrieval ratio. The "mask code" is based on the human face features and begin to work. This mask code can resist the light changes, skin tone, facial hair, hair, glasses, facial expression and posture, has strong reliability, so that it can from the million people accurately recognize someone. The recognition process of human face, use of common image processing equipment can be automatic, continuous, real-time [1

  [Return Home] [Print] [Go Back]   


Contact Us

  • Contact Person:

    DongGuang HongZhao Innovation Electronic Co., Ltd.
  • Tel:

  • Fax:

  • E-mail: