Single object face recognition technology research (1)

1 Introduction

As computer networks and communication technology, information security, intellectual property protection and authentication issues became an important and urgent issues.

Authentication is a necessary precondition for the security of the system, in a variety of different security areas require accurate identity authentication. Traditional card, smart card, passwords and other authentication methods exist to bring inconvenience, easily lost, unreadable or password easily cracked, and many other issues. Based on face recognition technology of identity authentication methods and traditional methods, have better security, reliability and validity, and thus is becoming more and more attention, and gradually into all spheres of social life.

Face recognition technology has a wide range of applications that can be applied to a variety of different security areas, because of its unique identification characteristics, uniqueness and relative stability, becoming a very popular research topics.

Many typical face recognition algorithms and applications are for standard or specific facial database, using the library for training in the face, and in the same library implementations face recognition. But in software protection, computer security, and other special applications, authentication only for a single object for face recognition, face recognition and incompetent such recognition task. Therefore, this article for a single object of face recognition features, discussed the single object of face detection and identification of key technologies, on the basis of a single object face recognition algorithms, experimental results proved the effectiveness of this method.

2 single object face recognition features

And the typical face recognition, face recognition of single object has the following four characteristics:

Application areas of the face recognition is a wide field of application, such as criminal cases and documents checking, security monitoring, and face recognition of single objects are mainly used in software protection, computer security lock, a specific object tracking, etc.

Recognition system by target single object face recognition with the ultimate goal of the system must have a high degree of security and reliability, i.e. the recognition error rates tend to be 0.

Although lowering the recognition error rates of recognition rate will be reduced, but you can prompt the user to adjust the attitude (such as eye camera, etc.).

Colour model as a single object face recognition only for a specific object, face detection colour model can adopt an adaptive method of adjusting the colour range.

Classification method for face recognition of single object does not exist, a common face database minimum distance classification is not able to correctly identify a particular object, you can only use threshold as a criterion.

As a result, the threshold selection is very important that the threshold is too large you prone to error, a security risk; whereas the threshold that is too small and will affect the recognition of efficiency.

3 face detection and normalization

Face detection is a prerequisite for face recognition.

For a given image, face detection to determine the image exists in the face, if exists, it returns its location and spatial distribution. Use of facial skin and facial features, will face detection is divided into two phases: outer face detection and face location. The main use of outside face detection face colour for initial face area detection, split out colour area; in the face detection is in the face area use facial geometry for verification and registration.

3.1-face detection

The mandate of the external face detection is to patiently image may face area to find out and marking of the steps are as follows:

(1) in accordance with human skin color in the color space in culture, features, and will possibly face pixel detection.

To make better use of colour characteristics and, at the choice of two HSI and YcbCr color space for image binarization, colour to the H is necessary [0, 46], S necessary [0.10, 0.72], Cb is necessary [98,130], Cr is necessary [128,170]. Will satisfy the condition of the pixel tag for colour pixels, and the rest are non-colour pixels.

(2) noise reduction processing.

In each colour point as the center of 5 × 5 o domain statistics number of colour pixels, more than half is the center point to color, otherwise considered non-colour.

(3) the binary image of colour blocks for regional integration, and the proportion of the target regions, structural analysis, filtering out impossible to face area.

Target range of the height/width ratio at 0.8 ~ 2.0.

3.2 face detection and positioning

Will contain the eyes, eyebrow, nose and mouth of the region known as the face region.

In the face area well expression facial features, and less vulnerable to such factors as the background, hair, as a result of interference in the face detection and positioning on the follow-up of feature extraction and recognition is vital.

In the face of the upper section of the binary image for both horizontal and vertical projection, to determine if two contain blackspots rectangle as the approximate area. binocular

In the determination of the two regions, on the black point for regional expansion, you can get a basic outline of the eye and the left-corner, black point coordinates of the location of the average value as a pupil.

Located around the pupil's coordinates are (Lx, Ly) and (Rx, Ry), two pupillary distance d between, on the basis of the human face of geometrical characteristics, we will face in the region is defined as: width =-d× 1.6, height =-d× 1.8 corner coordinates (Lx-d× 0.3, (Ly + Ry)/2-(-d) × 0.3).

Experiments show that the region can very well express facial features.

In the face of 3.3 normalized

[1] [2] [3]