Three-classification face detection method using context information

Abstract

The invention discloses a three-classification face detection method using context information. The method comprises the followings steps of 1) data preparation, 2) model design in which an end-to-endneural network model including a basic convolutional network, a regional proposal network and a fine adjustment network is designed; and 3) model training in which the neural network model designed in the step 2) is input via a training set, the model is trained in a batch random gradient decrease manner, a verification set is used to verify a training effect of the model, and finally a face detection model is obtained. The step 1) further comprises that a) each face in an image set is marked manually to obtain a face frame; b) the face frames after marking are classified according to the size, a small face area is amplified, and others are marked as normal faces; and c) a face image set with marks obtained in the step b) is divided into the training set and the verification set. Thus, the face detection precision and the face recall rate are improved greatly.

Description

technical field [0001] The invention relates to a human face detection method, in particular to a three-category human face detection method using context information. Background technique [0002] Face detection means that for any given image, a certain strategy is used to search it to determine whether it contains a human face (existence), and then return the position, size and posture of the human face. Face processing and analysis include face recognition, face tracking, pose estimation and expression recognition, among which face detection is the key first step in all face information processing. Most of the current face detection methods are based on the deep neural network framework. The main methods are: [0003] 1) Face detection based on cascaded convolutional neural network. Related patent: CN107688786A. Main technical means: Firstly, image preprocessing is carried out, and the test image is scaled and input to the first-level network. Secondly, in the subseq...

AI Technical Summary

Problems solved by technology

Problems and disadvantages: This type of method needs to set the minimum face parameters, and has a strong impact on the detection results; secondly, the network depth in the first stage is average, resulting in low face recall and a decrease in accuracy

Problems and disadvantages: The accuracy of extracting the foreground area from the picture to be checked by using the skin color segmentation method is not good, and the robustness to people with different skin colors is weak, which directly leads to inaccurate selection of samples based on the candidate area used for training; secondly, for the scale in the picture Smaller faces, poor detection ability

Images