| Semantic analysis of facial features and face replacement technology in privacy protection, intelligent human-computer interaction, facial image retrieval, video effects, and public safety has important application value. This paper presents a semantic-based automatic face replacement and solves the problem of glasses occlusion detection. Contribution of this paper includes the following two aspects:1) Propose and implement replace face method that based on semantic. According to the existing face recognition algorithms and target face the semantic description of the appearance of properties, we select the replace face of semantics"similar"or use a particular face. Firstly, preprocess the face image; include the face extraction, affine transform. Secondly, combined with facial features, use the SVM classifier to build predictive models that can automatically select the semantic similarity of replacing face.2) Implement replaces face that is obscured. Before making image fusion, we need to detect and locate in order to retain the part of the replacement (exist in the background) or replace the entire face region (the face region cover completely). Firstly, according to the selected samples, construct the corresponding glasses detection classifier in order to detect glasses based on the target face. Secondly, according to the result of obscured detect, select the appropriate replacement criteria, making the face after replacement realistic integration with the surrounding natural environment, complete replacement automatic in image and video.In the theoretical basis of the study, the authors designed and implemented an image / video frame of automatic face replacement. Experiments show that the proposed semantic-based face replacement and replacement under the conditions of the occlusion face will make the face replacement technology better, will directly promote the development of various practical applications, and has important theoretical and practical significance. |
PDF Full Text Request