Analysis And Research Of Human Face Similarity Features

With the rapid development of artificial intelligence and computer information technology,face recognition has gradually become a hot topic in the field of computer vision.The key technology of face recognition is to extract effective and stable identification features.The quality of the features directly determines the final face recognition effect.However,there are still many technical difficulties in how to locate and extract identification features which is stable and effective in the complex and changeable real environment such as lighting,noise and face occlusion.The main innovations and contributions of the paper are reflected in the following aspects:1.A similarity feature is analyzed and defined.Twins,multiple births or the same person at different ages have similar faces.This similarity is due to their genetic composition being similar or identical.Then whether there are some features related to genes in their faces?These features have great similarity and are less affected by the external environment,have certain stability,and have different composition ways in different faces.In this paper,these features are defined as similarity features.The similarity features of the face are unique and have better stability than other features of the face,and they are better recognized.2.According to the definition of similarity feature,a similarity feature extraction method based on Euclidean distance is proposed in this paper.First of all,all training samples are standardized,and each sample class is used as a whole to calculate the average value samples and standard deviation matrix of the sample class.Secondly,according to the size of the element values in the standard deviation matrix,the Euclidean distance between the training samples and the mean samples in each sample class is calculated in turn.Finally,in order to remove the redundant features with correlation in the samples,the sample features are retained when the Euclidean distance between the training samples and the mean samples reaches an extreme value.Experimental results show that this method can effectively extract face similarity features.3.This paper studies the combination of similarity features in different faces,and proposes a face recognition method based on weighted fusion of similarity features.Different frontal face images can be described by different linear combinations of similarity features.Firstly,the similarity feature is used as the basis,and the weight coefficients of the training samples in each type of samples on the basis are calculated,and a weight coefficient matrix is constructed.Secondly,the samples to be tested are projected on the basis of similarity features in the same way and the weight coefficients are calculated.Finally,the nearest-neighbor classifier is used to measure the similarity between the weight coefficient of the test sample on the base and the known weight coefficient,so as to achieve the classification of the face image.The experimental results in the self-built face database show that the method achieves a good recognition effect.