A face detection method and system based on convolutional neural network

Abstract

The embodiment of the present invention provides a face detection method based on convolutional neural network, including: inputting the picture to be detected into the improved single-step multi-scale detector ESFD after training, the ESFD is added on the basis of S3FD The enhanced interleaved structure and the context feature fusion structure are constructed after changing the anchor strategy and adjusting the loss function; the face detection result of the picture to be detected in the detection output layer of the ESFD is obtained. The embodiment of the present invention provides a face detection method and system based on a convolutional neural network. By modifying the network structure of S3FD, an enhanced interleaving structure and a context feature fusion module are added to the original S3FD network structure, and the anchor strategy is changed. And adjust the loss function to form a new ESFD architecture, which strengthens the ability to utilize features and improves the ability to detect small faces, thereby improving the recall and accuracy of face detection.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of face detection, in particular to a face detection method and system based on a convolutional neural network. Background technique [0002] In recent years, the SSD training framework has attracted much attention because of its fast detection speed and high accuracy in detecting large objects. [0003] In the existing technology, the face detection framework based on SSD is often used. The core of SSD is to use the convolution kernel on the feature map to predict the category scores and offsets of a series of default bounding boxes. In order to improve the detection accuracy, Predictions are made on feature maps of different scales. [0004] However, the existing SSD face detection framework is limited by the insufficient extraction of low-level features, which leads to its relatively poor performance in detecting small faces, resulting in low recall and accuracy of the entire detect...

AI Technical Summary

Problems solved by technology

[0004] However, the existing SSD face detection framework is limited by insufficient low-level feature extraction, which leads to its relatively poor performance in detecting small faces, resulting in low recall and accuracy of the entire detection. Therefore, a new method is urgently needed now. Face detection method to solve the above problems

Images