A Study On Automatic Generation Of Family Music Album Based On Emotion

With the development of the Internet and social network, electronic album has gradually become one of the basic services, and how to enhance the user experience of music album becomes particularly important. This thesis presents a study on automatic generation of family music album based on emotion, which is designed to solve the problems of the photo album recommendation for a music that you like based on the same emotion. According to the emotions in the musics and images, we selected the representative features, using the canonical correlation analysis(CCA) to study the relevence between the musics and images in the same emotion. Finally we have put forward the basic mapping model and realized the automatic generation for the music album.In this thesis, according to the emotional classification standard of MIREX, emotions can be divided into five kinds: passionate, cheerful, literate, witty, volatile. On the basis of Thayer's two-dimensional emotion model, the V(Valence)?A(Arousal) value is used to determine the emotional coordinates of musics and images, and to define the emotional range s. This paper selects MFCC and RASTA-PLP features in musics, and color features for images. In the algorithm, the canonical correlation analysis(CCA) is used to analyze the relavence between the two sets of variables. This algorithm solves the heterogeneous problem of image features and music features, and realizes the cross modal fusion.In this thesis, the CCA method is compared with some other methods: random recommendation and manual recommendation, the results show that the CCA method has got the 67.45% satisfaction in mannual evaluation, which is close to the results 78.09% of the manual recommendation. In comparison with the testing music in known and unknown emotion, it is found that CCA method has good effectiveness in the correlation analysis between the two variables.