| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 102 === With the popularization of the digital devices, more people like to share their personal videos on the Internet. Before uploading these videos, people may want to add proper music segments to the captured videos. But such the music adding tasks are all performed by manual processes. In this thesis, we proposed an emotion-based optimized music adding system, which can analyze a video’s emotions and find the best arrangement of music clips for the video from the music database by means of an optimization process.
The proposed system estimates the video emotion curve using a radial basis neural network, and then applied the proposed Flexible Harmony Search Algorithm (FHSA) to find the best music arrangement whose emotional transition is most similar to what of the video. FHSA is composed by two layers. The first layer determines the length of each clip for a music arrangement and the second layer selects the music clips to fill the arrangement. This system finds the global optimized music arrangement by iteratively generating new solutions, comparing candidate solutions and updating the selected solutions.
The contributions of this proposed work are: (1) The proposed work is the first one to incorporate a heuristic optimization algorithm into the music adding system. (2) Unlike HSA, the proposed FHSA breaks the fixed-parameter limitation and is able to get the optimized music arrangement without the constraints on the length and number of music clips. The experimental results show that the convergence speed of FHSA is 8 times faster than modified HSA with similar fitness values. The subjective evaluation also demonstrates that the video with added music segment can effectively enhance the video emotion.
|
|---|
