Heterogeneous Face Recognition Based On Composite Sketch

Sketch face recognition refers to identifying the corresponding identity of the sketch image by comparing the sketch face image in the face photo library,which is mainly used in the field of looking for the identity of the suspect.Compared with traditional face recognition for the same modality,sketch face recognition aims to solve the difference between different modalities,and its application range is wider.Composite sketch face recognition is a branch of sketch face recognition.Compared with other types of sketches,composite sketch is popular in the field of criminal investigation due to its characteristics of fast image synthesis and low cost.Simultaneously,more and more researchers are paying attention to it.At present,although the existing composite sketch face recognition algorithm has achieved some good results,there are still many problems to be overcome.This paper researches on the single feature model and the weight determination in feature fusion problems of existing algorithms,and its main contributions are as follows:(1)To solve the problem of single feature model in existing algorithms,this paper proposes a composite sketch face recognition method based on multi-scale low-level features and semantic attributes.This method extracts the global contour features and local detail features of the image,and fuses the extracted features on score level through a weighted fusion method to obtain their corresponding matching results.Afterwards,the high-level semantic attribute features are used to filter the matching results obtained by the above multi-scale low-level feature model to obtain the final matching result.The novelty of this method is:solving the problem of the single feature model through extracting the multi-scale low-level features and high-level semantic features.(2)To solve the problems of single feature model and the weight determination in the feature fusion,this paper proposes a multi-level feature fusion recognition algorithm model based on adaptive algorithm.This method describes the low-level feature by extracting the contour features and detail features of the human face,and extracts the deep facial feature by using the pre-trained model of the deep learning method to represent the high-level features of the image,which can solve the problem of single feature model.Afterwards,the weight of each feature is determined by the adaptive weight method proposed,and each feature is weighted and fused on the score level,which can solve the problem of weight determination in the feature fusion.