CN111368803A - A face recognition method and system - Google Patents

Abstract

The invention discloses a face recognition method and a system, wherein the system comprises the following steps: the face target detection module is used for mining a face target in the image based on a pyramidBox algorithm; the feature extraction module is used for extracting LBPH, SIFT and human face skeleton features of the human face target; the feature fusion module is used for fusing the extracted human face bone features, SIFT and LBPH3 features according to the weight of 3: 1: 5 to obtain a human face feature vector; and the face recognition module is used for recognizing the face by adopting a nearest neighbor classifier based on the face feature vector. The invention uses the fusion of various characteristics to recognize the face, can effectively improve the influence of single characteristics on the recognition due to the illumination, the angle and the scale change, improves the recognition rate, realizes the automatic locking of the unrecognized face, thereby reducing the recognition loophole and being convenient for a user to lock a suspicious object quickly.

Description

A face recognition method and system

technical field

The invention relates to a face recognition system, in particular to a face recognition method and system.

Background technique

Face recognition is a biometric technology for identity authentication based on human facial feature information. Capture images or video streams containing faces through cameras or cameras, and automatically detect and track faces in the images, so as to match and identify the detected faces.

The application field of face recognition is very wide, especially in many fields such as security and anti-terrorism, financial payment, access control attendance, identity recognition, etc. It plays a very important role. study etc.

At present, face recognition algorithms mainly include: (1) Template matching method: It mainly uses the texture and grayscale features of the face, and uses the nearest neighbor classifier.

Match the image to be recognized with all templates in the dataset to find the most similar template. The main problem with this method is that if the data set is large enough, it will consume a lot of time when matching, and the accuracy will be reduced.

(2) Principal Component Analysis (PCA): In short, the image of the original image library is represented by a low-dimensional feature, which is generally required to represent more than 90% of the entire image, and can be used to a large extent. reduce the amount of computation. However, the algorithm is sensitive to the external environment, and in the process of identification, the number of initial features is difficult to determine in advance.

(3) Support Vector Machine (SVM): SVM is a powerful classifier. In general, the algorithm using this classifier has a higher accuracy. When the number of face samples is large, the time complexity and space complexity of the algorithm are high.

(4) Linear Discriminant Analysis (LDA): The basic idea is to project the high-dimensional pattern samples into the optimal discriminant vector space to achieve the effect of extracting classification information and compressing the dimension of the feature space. The subspace of has the largest inter-class distance and the smallest intra-class distance. Overemphasizing the large class spacing and ignoring the features with small class spacing will eventually cause a large number of overlapping categories with small class spacing, resulting in a low final recognition accuracy.

SUMMARY OF THE INVENTION

In order to improve the problem of low recognition rate caused by the face recognition method based on only one feature and a single recognition method, the present invention provides a face recognition method and system, which uses multiple feature fusion for face recognition, which can effectively improve the single recognition rate. The feature improves the recognition rate due to the influence of illumination, angle and scale changes on the recognition.

To achieve the above object, the technical scheme adopted in the present invention is:

A face recognition system, comprising:

The face target detection module, based on the PyramidBox algorithm, realizes the mining of face targets in the image;

a feature extraction module for extracting the LBPH, SIFT and face skeleton features of the face target;

The feature fusion module fuses the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain a face feature vector;

The face recognition module uses the nearest neighbor classifier to realize face recognition based on the face feature vector.

Further, the facial skeleton features are extracted based on the Kinect depth sensor.

Further, it also includes:

The face target hiding module is used to hide the face target that has completed face recognition in the image.

Further, it also includes:

The detection module for glasses, masks and hats, based on the ssd_inception_v2_coco model, realizes the detection of glasses, masks and hats in the images processed by the face target hiding module;

The glasses, mask and hat delineation module is used to mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, mask and hat detection module.

Further, the ssd_inception_v2_coco adopts the ssd target detection algorithm, pre-trains the inception v2 deep neural network with the coco data set, and then trains the model with the previously prepared data set, fine-tunes the parameters in the deep neural network, and finally obtains a suitable Object detection models for glasses, masks, hats.

The present invention also provides a face recognition method, which is implemented based on the above-mentioned face recognition system and includes the following steps:

S1. Based on the PyramidBox algorithm to realize the mining of face objects in the target image;

S2, extracting the LBPH, SIFT and face skeleton features of the face target;

S3, fuse the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain face features;

S4, using the nearest neighbor classifier to recognize the face based on the face feature.

Further, it also includes the following steps:

S5, based on the face target hiding module to realize the hiding of the face target that has completed face recognition in the image;

S6. Based on the ssd_inception_v2_coco model, the detection of glasses, masks and hats in the image processed by the face target hidden module is realized;

S7. Mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, masks and hats;

S8 , observe, compare and track the objects in the marked position frame manually.

The present invention has the following beneficial effects:

1) Efficient mining of face targets can be achieved based on the PyramidBox algorithm, and face recognition based on the fusion of LBPH, SIFT and face skeleton features can effectively improve the influence of a single feature on recognition due to changes in illumination, angle and scale, and improve the recognition rate. .

2) The automatic locking of unrecognized faces is realized, which can reduce identification loopholes and facilitate users to quickly lock suspicious objects.

Description of drawings

FIG. 1 is a system block diagram of a face recognition system according to an embodiment of the present invention.

FIG. 2 is a flowchart of a face recognition method in Embodiment 2 of the present invention.

FIG. 3 is a flowchart of a face recognition method in Embodiment 3 of the present invention

Detailed ways

In order to make the objects and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

Example 1

As shown in FIG. 1, an embodiment of the present invention provides a face recognition system, including:

The face target detection module, based on the PyramidBox algorithm, realizes the mining of face targets in the image;

A feature extraction module for extracting the LBPH, SIFT and face skeleton features of the face target; the face skeleton features are extracted based on the Kinect depth sensor;

The feature fusion module fuses the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain a face feature vector;

The face recognition module uses the nearest neighbor classifier to realize face recognition based on the face feature vector;

The face target hiding module is used to hide the face target that has completed face recognition in the image;

The detection module for glasses, masks and hats, based on the ssd_inception_v2_coco model, realizes the detection of glasses, masks and hats in the images processed by the face target hiding module; the ssd_inception_v2_coco adopts the ssd target detection algorithm and pre-trains the inception v2 deep neural network with the coco data set , and then use the previously prepared dataset to train the model, fine-tune the parameters in the deep neural network, and finally obtain a suitable target detection model for glasses, masks, and hats;

The glasses, mask and hat delineation module is used to mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, mask and hat detection module.

Example 2

As shown in FIG. 2 , an embodiment of the present invention further provides a face recognition method, including the following steps:

S1. Based on the PyramidBox algorithm to realize the mining of face objects in the target image;

S2, extracting the LBPH, SIFT and face skeleton features of the face target;

S3, fuse the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain face features;

S4, using the nearest neighbor classifier to recognize the face based on the face feature.

Example 3

As shown in FIG. 3 , an embodiment of the present invention further provides a face recognition method, including the following steps:

S1. Based on the PyramidBox algorithm to realize the mining of face objects in the target image;

S2, extracting the LBPH, SIFT and face skeleton features of the face target;

S3, fuse the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain face features;

S4, using the nearest neighbor classifier based on the face feature to realize face recognition;

S5, based on the face target hiding module to realize the hiding of the face target that has completed face recognition in the image;

S6. Based on the ssd_inception_v2_coco model, the detection of glasses, masks and hats in the images processed by the face target hidden module is realized; the ssd_inception_v2_coco adopts the ssd target detection algorithm, pre-trains the inception v2 deep neural network with the coco data set, and then uses the previously prepared The model is trained on the data set, fine-tune the parameters in the deep neural network, and finally obtain a suitable target detection model for glasses, masks, and hats;

S7. Mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, masks and hats;

S8 , observe, compare and track the objects in the marked position frame manually.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be It is regarded as the protection scope of the present invention.

Claims (7)

1. a face recognition system, is characterized in that, comprises: The face target detection module, based on the PyramidBox algorithm, realizes the mining of face targets in the image; a feature extraction module for extracting the LBPH, SIFT and face skeleton features of the face target; The feature fusion module fuses the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain a face feature vector; The face recognition module uses the nearest neighbor classifier to realize face recognition based on the face feature vector. 2 . The face recognition system according to claim 1 , wherein the skeleton features of the face are extracted based on a Kinect depth sensor. 3 . 3. a kind of face recognition system according to claim 1, is characterized in that: also comprise: The face target hiding module is used to hide the face target that has completed face recognition in the image. 4. a kind of face recognition system according to claim 1, is characterized in that: also comprises: The detection module for glasses, masks and hats, based on the ssd_inception_v2_coco model, realizes the detection of glasses, masks and hats in the images processed by the face target hiding module; The glasses, mask and hat delineation module is used to mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, mask and hat detection module. 5. A kind of face recognition system according to claim 4, it is characterized in that: described ssd_inception_v2_coco adopts ssd target detection algorithm, pre-trains inception v2 deep neural network with coco data set, and then trains with previously prepared data set This model fine-tunes various parameters in the deep neural network, and finally obtains a suitable target detection model for glasses, masks, and hats. 6. a face recognition method, is characterized in that: realize based on the described face recognition system of any one of claim 1-5, comprise the steps: S1. Based on the PyramidBox algorithm to realize the mining of face objects in the target image; S2, extracting the LBPH, SIFT and face skeleton features of the face target; S3, fuse the extracted face skeleton features, SIFT and LBPH features with a weight of 3:1:5 to obtain face features; S4, using the nearest neighbor classifier to recognize the face based on the face feature. 7. a kind of face recognition method as claimed in claim 6 is characterized in that: also comprise the steps: S5, based on the face target hiding module to realize the hiding of the face target that has completed face recognition in the image; S6. Based on the ssd_inception_v2_coco model, the detection of glasses, masks and hats in the image processed by the face target hidden module is realized; S7. Mark the position frame in the image processed by the face target hiding module according to the detection results of the glasses, masks and hats; S8, observe, compare and track the objects in the marked position frame manually.

Images