The Research Of Face Ethnicity Recognition Based On Deep Learning

Since 2006, Along with the paper that Hinton wrote about deep learning published in Science, a new wave of deep learning has arisen. The ability of self-learning features makes deep learning useful in the area of theoretical study and industrial production. For now, deep learning is probably the closest artificial intelligence to human brain.Through the constant exploration,scientists around the world has put up many models about deep learning,such as AutoEncoder, RBMS,CNNS,etc. Among them, CNNS has been used in image recognition,speech recognition,natural language processing and other fields,and it has achieved a lot.Therefore, many apps has arisen at the moment.In our country, a multiracial one, the study about characteristics of faces in different nations,can not only help us understand the process of reproduction and evolution in different nations, but also save the facial features in different nations by means of technology, and preserve a valuable wealth for the future study of ethnology and anthropology. Face ethnicity recognition can expand and enrich the face detection and recognition fields at the same time.This very article has put up our ethnicity face database,ChineseEthnicityFace. Firstly, I use the traditional PCA to extract face algebraic features and adopt knn to make classification recognition.Afterwards, I compare the performances of different CNNS in ethnicity faces.Then, based on CNNS and multiple structure differentiation integration and ensemble learning,I propose an ensemble FF_CNNS. And the experiment result shows that the accuracy with FF_CNNS is higher than single CNNS.Considering the complexity of collecting large-scale ethnicity face in reality, this article adopts the way of web crawler to get pictures of ethnicity faces,and build the database.By comparing the different recognition abilities in face from ethnic people with different CNNS,such as T_CNNS?F_CNNS?FV_CNNS. At last,I choose multiple F_CNNS and FV_CNNS to build a differential integration CNN model,FF_CNNS. Finally I design and implement a face ethnicity video recognition system,and obtain good accuracy. This is the very first time that CNNS and integrated learning are applied to the face ethnicity recognition.