Spatial coherence feature-based quick identification method for human face expression of any pose

Abstract

The invention discloses a spatial coherence feature-based quick identification method for a human face expression of any pose. The method comprises the steps of firstly, synthesizing a front face image corresponding to a human face image of any pose; secondly, detecting 51 key feature points based on the synthesized front face image, and extracting key regions by taking the feature points as centers; thirdly, performing quick unsupervised feature learning based on the key regions; and finally, performing convolution sum pooling by taking each key region as a unit to obtain unsupervised feature learning-based high-level features, obtaining a spatial coherence feature used for identifying the human face expression of any pose in combination with the high-level feature and a geometric position feature of each key region, and inputting the spatial coherence feature to an SVM for training to obtain a unified expression identification model, thereby finishing the identification of the human face expression of any pose. According to the method, the problems of low identification rate due to the fact that conventional features do not have a spatial constraint relationship, low efficiency due to the fact that a model needs to be built for each pose in conventional multi-pose human face expression identification, and the like are solved, so that the identification accuracy and efficiency are effectively improved.

Description

technical field [0001] The invention belongs to the field of emotion recognition, and in particular relates to a method and system for fast recognition of human facial expressions in arbitrary poses based on spatial coherence features. Background technique [0002] Facial expression recognition is an important research direction in the fields of pattern recognition, human-computer interaction and computer vision, and has become a research hotspot at home and abroad. Generally speaking, the six most common basic human expressions are happiness, sadness, anger, surprise, disgust and fear. In recent years, with the continuous proposal of various features that are robust to poses, the development of multi-pose automatic facial expression recognition technology has been promoted. For example, the traditional face recognition model can only perform facial expression recognition based on pictures of frontal or near-frontal faces, but the recognition effect on facial expressions of...

AI Technical Summary

Problems solved by technology

[0003] (1) In the paper of S.Eleftheriadi et al., entitled "Discriminative Shared Gaussian Processes for Multiview and View-Invariant Facial Expression Recognition", a discriminative Gaussian process hidden variable model is used for multi-pose facial expression recognition, but This method uses traditional manual features, which are not robust to object occlusion, face deformation and posture changes.

(2) In the paper titled "Joint Fine-Tuning in Deep Neural Networks for FacialExpression Recognition" by H.Jung et al., the face features learned by cascaded convolutional neural network and the geometric features of key areas of the face The method has obtained good expression recognition results, but the method is based on a complete face when learning features through the convolutional neural network, which makes the combined features learned by it different from those learned by the feature learning method. Has spatially coherent features, which reduces the recognition rate of the model

Images