Dynamic sequence unconstrained expression recognition method based on mixed feature enhancement network

Abstract

The invention discloses a dynamic sequence unconstrained expression recognition method based on a mixed feature enhancement network, and relates to the technical field of facial expression recognition. The method comprises the steps of 1, carrying out the human face detection of human face expression video data, intercepting a human face ROI region, and removing background interference, and obtaining dynamic sequence human face expression data; 2, dividing the dynamic sequence human face expression data into multiple groups of sequences by taking N frames as one group, analyzing the multiple groups of sequences, extracting expression features of the sequences, and overlapping N / 2 frames of images between each group of sequences; and 3, inputting N frames of images of each group into the single-frame feature enhancement CNN network and the multi-frame feature enhancement self-attention network in sequence to obtain N 2048-dimensional feature vectors. According to the invention, the discrimination capability of dynamic sequence facial expression features is effectively improved, and the intra-class gap of unconstrained expression data is reduced; and a shallow feature enhancement module provided by the invention improves the discrimination capability of expression features by increasing the network width, and effectively reduces the calculation complexity.

Description

technical field [0001] The invention relates to the technical field of facial expression recognition, in particular to a dynamic sequence unconstrained expression recognition method based on a mixed feature enhancement network. Background technique [0002] In recent years, as the demand for facial expression recognition in practical applications has become more and more extensive, the research objects have gradually shifted from laboratory-constrained facial expression recognition to real unconstrained facial expression recognition. From continuous exaggerated expressions to instantaneous micro-expressions, a huge change from basic expression classification to complex expression analysis. As a result, the traditional facial expression recognition methods are no longer competent, and the deep unconstrained expression recognition network with strong learning ability has developed rapidly and achieved remarkable results. [0003] The existing deep unconstrained facial express...

AI Technical Summary

Problems solved by technology

[0004] However, in practical applications, due to the complexity of real spontaneous expressions, the differences in individual attributes such as age, gender, and race, and the influence of unconstrained environments such as lighting, posture, and occlusion, the amount of facial dynamic expression data collected is small and the data standards are difficult. , The quality is uneven, these factors lead to the existing dynamic unconstrained expression recognition system is far from reaching the practical level, and there is still a lot of room for improvement in system performance

Images