- 學位論文
從影片中自動擷取生成與預設人體姿勢相符之圖片
Frame Retrieval and Synthesis with Predefined Hand-Drawn Shape from Videos
摘要
由於數位相機和手持裝置的普及,照相越來越簡單容易,且已經成 為人們日常生活中普遍的活動,人們對於拍照的姿勢也越來越多變。 近年來,不論是為了藝術目的或是其他特殊需求,人們喜歡拍一些靜 止於空中或是跳躍的照片,不過要拍這種照片通常會遇到一些麻煩, 像是場景中每個人跳躍的時間點不同、或是攝影師按下快門的時間不 適合,都會造成拍出的照片與所想像的姿勢不符的情形,因而浪費許 多時間。 本篇論文提出一個簡單拍出跳躍照的架構,使用者只須隨手畫出想 要拍出的姿勢草圖,系統便會生成出與所畫姿勢相符之圖片。整個架 構首先取出物體的形狀,對其做骨架化,並將之與所繪形狀之骨架做 比對,將擁有想要姿勢的影像從影片中擷取出來,最後把符合期待的 每個人的姿勢做結合,生成出與所繪相符之圖片。 實驗結果顯示基於骨架之比對方法能準確擷取出與所繪形狀最相似 之圖片,並且能實際應用在拍攝特殊跳躍照之情況。
並列摘要
In recent year, people like to take the jumping photos or pictures which are human freezing in the air for artistic purpose or other demands. When taking these kind of photos, some problems occur, such as inconsistent jumping time, or a improper shooting moment which would make the photo undesirable. These problems waste much time for taking an available picture. In this thesis, we propose a framework for taking these kind of photos easily. Given a hand-drawn posture sketch, the system would synthesize the desirable image according to the sketch. The system extracts out object shape for each frame at first. Then it computes the skeleton of the shape. By computing the similarity between two skeletons of extracted shape and hand-drawn shape, the frame with desirable posture would be retrieved from the video. Finally, the system synthesize the image by composing every desirable postures. The experimental results shown that the matching method based on skeleton can retrieve the frame which is most similar to hand-drawn shape. The system is applicable to the condition of taking jumping photo.