以情緒為基礎的演化式影音配對系統

Abstract

近年來,由於高品質的相機,影片編輯軟體,以及音樂編曲軟體變 得愈來愈容易取得,製作影片從一項專業性的工作變成非常普遍的活 動。然而,處理家庭錄製的影片往往會耗上大量的時間。許多被廣泛 使用的商業軟體例如像是iMovie與Adobe Premiere等提供了對使用者非 常友善的介面,使得人們可以自行製作喜愛的影片,然而要幫影片加 上合適的音樂卻仍然不是一件容易的事,大多數的時候這項工作相當 依賴人類的感覺與情感。在這篇論文裡面,我們提出了一個依據情緒 內容來進行影音配對的系統,學習的部份則採用實數基因演算法來達 成。實驗結果顯示我們的系統能夠根據已知的配對資料去產生合適的 影音組合,因此這個系統能夠根據學習的資訊去推薦使用者合適的音 樂清單,進而達到加速影片製作流程的效果。

As the high quality cameras, video editing software and music composi- tion software are more available, making videos becomes popular. However, processing home videos is time-consuming. Some popular commercial soft- ware such as iMovie and Adobe Premiere provide user-friendly interface to help people make films on their own, but adding adequate music to a clip is still a non-trivial work that highly depends on human feelings and emotions. We propose an emotion-based evolutionary music video pairing system by utilizing affective information of videos and musics to remedy this problem in this thesis. Empirical results show that our system is capable of pairing videos with adequate musics and self-adapting to human preferences. It is expected to accelerate the process of producing attractive videos by reducing the efforts needed in searching appropriate musics for videos.