Expression Recognition Method Based On Neighborhood Systems

Expression recognition is an important research direction in the field of artificial intelligence,and its research methods mainly include data processing,expression feature extraction and expression classification.How to extract expression features quickly and efficiently is a challenging research topic.The main methods of expression recognition include traditional feature-based expression recognition methods and recognition methods using deep neural networks,which have their own advantages and disadvantages.This paper attempts to combine the advantages of the two,learn from human cognitive processes to form a set of human-like intelligence algorithms,and use computers to achieve face expression recognition.In the process of combining the traditional expression recognition method and the deep neural network recognition method,the quality of data processing directly affects the quality of feature extraction.Therefore,this article makes two attempts at data processing.The first is data cleaning,the original data through the face landmark to locate the face area,as well as the most affected expression of the eyebrow area and mouth and nose area,and under the theoretical support of the face coding system,through the experiment to find a suitable feature neighborhood about the eyebrow and nose,the neighborhood has the greatest impact on the accuracy of face expression.The second is data fusion,in order to allow an image to contain as much useful information as possible,the three key regions after data cleaning are fused to form an image that contains three key areas at the same time.Subsequently,based on the idea of neighborhood system,three concepts are proposed,namely square neighborhood,characteristic neighborhood and co-occurrence correlation features with causal relationship,and at the same time,it is theoretically proved that the more co-existing correlation features detected,the higher the recognition accuracy,and experimentally verifies that the network in this paper is similar to the human cognitive process to a certain extent.In order to verify the rationality and feasibility of the theory in this paper,a multi-region integrated network is first built to compensate for the problem of single network feature extraction.In order to verify the rationality and feasibility of the theory in this paper,a multi-region integrated network is first built to compensate for the problem of single network feature extraction.The integrated network can extract the facial features,eyebrow features,and mouth and nose features at the same time,assign different weights to the three networks according to their different contributions,and finally use the trained multi-area integrated network to identify the test samples.Experimental results on the classical expression dataset CK+ and the mixed expression dataset CJKR show that the proposed method can effectively improve the expression recognition accuracy,but also increase the network complexity.In order to reduce the complexity of the network while maintaining a high accuracy,an expression recognition method based on neighborhood system is proposed,which feeds the images that have been fused by data into a single neural network,so that the network pays more attention to important areas related to expressions.On two classical expression datasets,the experimental data obtained by using the proposed method show that the proposed method is simple and effective,and compared with a single network,the proposed method can extract the features of three different perspectives at the same time,effectively improving the accuracy of expression recognition.This method is simpler and easier than an integrated network.