Home /

Key Technology

/Facial Recognization
Facial Recognization
Face Recognition Alqorithmsa

With the abilily to develop profesional visible light module and dual camera module, providing professional level office attendance, access control management and other office product development by using Baidu face recognition algorithm. Courseware Module Supporting Strong Backlighting: Monocular and Binocular Face Algoithms: Baidu







Artificial Intelligence Binocular Recognition









Using infrared LED lamp, in any light complex environment, near infrared camera can provide accurate image for face recognition algorithm; Baidu algorithm can quickly realize the determination of living body.




Ultra-wide dynamic visible light module provides a higher contrast and high definition image, which can be recognized quickly even under strong backlight conditions through Baidu algorithm.


















Image Preprocessing






Grayscale


There are three methods to transform color image into gray image: maximum method, average method and weighted average method.



Geometric Transformation


The collected images are processed through geometric transformations such as translation,  transposition,  mirror, rotation and zoom in order to correct the systematic errors from the image acquisition system.













Image Enhancement


The purpose of image enhancement is to improve the quality of face image, make the image clearer in vision, and make it easier to recognize.



Normalization


The goal of normalization is to normalize face images with the same size and gray scale.













Deep Learning.Feature Points




01


02


03


04




Feature Detection. Algorithms


On the basis of face detection, key feature points such as eyes, nose tips, mouth corners, eyebrows and contour points of various parts on face can be automatically located according to the input face images.

Input: Face Appearance
Input: Face Appearance



Face Alignment. Algorithms


To search for pre-defined points (also known as face shape) in a face image, it usually starts with a rough shape, and then refines the shape estimation by iteration.

Methods include:
1.AAM(Active Appeamce Model)
2.ASM(Active Shape Model)


Feature Point Location. Algorihms


Using CNN, the key points of face can be located accurately from coarse to fine. The general network structure is divided into three layers: level 1, level 2 and level 3. Each layer contains several independent CNN models.


Feature Point Extraction. Algorithms


A predictor is needed to construct a feature extractor to train the model. To download the trained key point extraction model provided by dlib.
















Face Recognition












Steps

Principle

Calculating Euclidean Distance



1. Import model parameters first

2. Then two pictures are imported to obtain the 128-dimensional feature vectors from the model.
3. Finally, calculated the Euclidean distance between two vectors.



The Euclidean distance between feature vectors is calculated to get the similarity of face.

The recognition rate of LFW is 99.63%.


Euclidean metric (also known as Euclidean distance) is a commonly used definition of distance, which refers to the real distance between two points in m-dimensional space, or the natural length of a vector (the distance between the point and the origin). Euclidean distance in two-dimensional and three-dimensional space is the actual distance between two points.









Related Products
H1 H1
H2 H2
H4 H4
Leave a message
Leave a message
If you are interested in our products and want to know more details,please leave a message here,we will reply you as soon as we can.

Home

Products

about

contact