Face detection method based on hierarchical network and cluster merging

Abstract

The invention discloses a face detection method based on hierarchical network and cluster merging. The face detection method based on hierarchical network and cluster merging divides the convolutionalneural network into two-level of networks. In the first-level network training process, by obtaining heat maps of the original input picture at the seven-level resolution, by obtaining the initial candidate face area at the seven-level resolution according to the local hottest area on the seven heat maps, loss of face information on the sample can be reduced. In the second-level network trainingprocess, the specific facial feature area is added in the training positive sample, so that the convolutional neural network can extract the features of the facial feature area in a targeted manner, and at the end of the network, a picture containing a face detection frame is obtained through a candidate box merging strategy based on cluster and facial features. The face detection method based onhierarchical network and cluster merging overcomes the problems that a current method is complicated in network and large in calculated amount, effectively processes the face which changes the postureor is shielded in the picture, and improves the face detection accuracy based on the convolutional neural network.

Description

technical field [0001] The invention relates to the field of face recognition, in particular to a face detection method based on hierarchical network and clustering merging. Background technique [0002] Face detection is the basis of various face analysis tasks, and its detection accuracy directly affects the performance of subsequent tasks. However, in actual scenes, due to the interference of external factors, such as illumination, occlusion, and changes in human expression and posture, face detection has always been a challenging problem in face analysis tasks. [0003] Since the convolutional neural network model trained with deep learning algorithms was proposed, it has achieved remarkable results in multiple large-scale recognition tasks in the field of computer vision, and has become a research hotspot in recent years. In the existing face detection methods based on convolutional neural networks, Bootstrapping Face Detection with Hard NegativeExamples uses a deep 50...

AI Technical Summary

Problems solved by technology

In the existing face detection methods based on convolutional neural networks, Bootstrapping Face Detection with Hard NegativeExamples uses a deep 50-layer residual network to achieve the purpose of improving detection accuracy. Although the method has achieved good results, but The network design is too complex and the amount of calculation is too large

Object Specific Deep Learning Feature and Its Application to Face Detection uses a multi-resolution sliding window to process small faces in the picture. Although it has achieved good detection results, this method combines the heat maps of the obtained multi-level resolution pictures in advance , will cause a certain loss of face information, resulting in a decrease in detection accuracy

In addition, existing face detection methods based on convolutional neural networks are still relatively low in robustness to pose changes and occluded faces.

Images