A face detection method under unrestricted conditions

Abstract

The invention provides a face detection method under a non-limiting condition. The face detection method comprises the following steps of S1) preprocessing an image; S2) designing a face detection network based on deep convolution; S3) carrying out forward propagation of the face detection network; S4) adopting a non-maximum suppression algorithm; S5) obtaining a final detection result. The methodhas the advantages that the application range is wide, the effect and the speed can reach the state-of-art level, the accuracy of pedestrian coordinates can be improved, the false detection occurrence probability can be reduced, the deep network gradient dispersion problem can be relieved, and the network convergence process can be accelerated.

Description

technical field [0001] The invention relates to the technical field of computer image processing, in particular to a face detection method based on a deep convolutional neural network combined with a multi-scale feature pyramid under unrestricted conditions. Background technique [0002] As the basis of various visual tasks, face detection technology occupies a very important position in the fields of image processing and pattern recognition. In recent years, with the rapid development of artificial intelligence based on neural networks, face detection technology has been increasingly used in various visual tasks, such as witness comparison, meeting sign-in, face gate, face recognition The premise of such tasks is a high-precision, high-accuracy face detection method. [0003] Early face detection technology relied on manually constructed features, combined with traditional machine learning. For example, the famous Haar feature and the face detection algorithm using the Ad...

AI Technical Summary

Problems solved by technology

Among them, MTCNN uses small neural network classification and sliding window for fast detection, which can achieve good detection effect and speed, but its generalization performance is poor, and it needs to be retrained for specific scenarios; YOLO and SSD use deep convolutional networks. The anchor points of each feature map are classified and biased by regression at one time to realize the detection process. This type of method is faster in speed and better in generalization performance, but has a loss in accuracy; the performance of the second-stage Faster RCNN algorithm is relatively The other types are the best, but due to the intervention of the fully connected layer, the amount of calculation is greatly increased, and it is difficult to apply to industrial application scenarios that require real-time performance

Images