A Name Recommendation Mechanism Using Modified Probability Neural Network

• Main Authors: Shiang En Tang, 唐祥恩
• Other Authors: Ding Horng Chen
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/36225521323894481551

碩士 === 南台科技大學 === 資訊工程系 === 97 === The internet changes the behavior of peoples. The community web site can share multimedia files and thus increase the usage of photo album. Among the application of photos, the human face is the most significant feature for recognizing a specific person. Therefore, face detection is widely used in various applications. If the system can annotate a face with a proper name tag, the linkage will allow user to search the photo album not only by keyword, but also by their real name. In this thesis, we have developed a web photo album with face recognition, tag annotation and name recommendation functions. For each picture in the photo album, the system will automatically detect and locate the face regions in the picture, and then compute the face features and analyze the characteristics store in the image database. Users can add a name tag or an annotation text for each face region. If the face amount is large enough to complete the training phase, the system will build an association between the new adding face and the name tag in the face database via the probability neural network. This system will also provide a convenient interface to annotate the face if the face recognition result is not so perfect. The experiment indicates that the modified probability neural network can correctly train the annotated face images. For a newly added face image, the system will recommend a proper name to this image. But we also find that the misclassification will occur for some ambiguous faces. In the future, we will try to enhance the recommendation performance with much refined feature processing and user feedback mechanism.