Facial expression analysis method based on capsule network

Abstract

The invention provides a facial expression analysis method based on a capsule network, and the method employs a multi-channel parallel technology, a Strict-Squash function, a Lane-Filter function, and DropCircuit technology to improve an original capsule network, avoids noise interference caused by the irrelevant features of facial expressions, reduces the workload of manual marking, and improves the accuracy of facial expression analysis. Therefore, sentiment analysis can be carried out on a face image with a complex background, and the expression recognition precision is improved. The method is innovated on the basis of the capsule network, can effectively learn the spatial relationship between objects, has a strong recognition capability for slightly deformed objects and pictures of the same object at different visual angles, is not liable to be interfered by noise caused by irrelevant features of facial expressions, and has a good recognition effect on the objects at different visual angles. The method can carry out correct sentiment analysis on a face image with a complex background, is high in expression recognition precision, is wide in application range, and is high in robustness.

Description

technical field [0001] The invention relates to the fields of facial expression recognition, emotion analysis and deep learning. Background technique [0002] When collecting face images, it is inevitable to introduce noise such as lighting and background; and there are individual differences in the face samples themselves, specifically in age, gender, skin color, fat and thin, whether to wear glasses, etc. These noises and differences will weaken the characteristics of expression information Expression, reducing the accuracy of expression recognition, directly inputting original pictures to extract expression features will cause a great negative effect on the recognition results. Secondly, the existing facial expression database has a small amount of data, which is difficult to meet the training needs of the deep network, and the obtained deep model is prone to overfitting problems. [0003] Facial expression recognition refers to the analysis of a given face image or imag...

AI Technical Summary

Problems solved by technology

However, the current capsule network technology is susceptible to noise interference caused by features irrelevant to facial expressions, which increases the workload of manual labeling, making it difficult to perform emotional analysis on face images with complex backgrounds, and the accuracy of expression recognition is not high.

Images