光影塗鴉是利用相機長曝光去記錄光源運動的軌跡所形成之照片, 藉由此種技術,我們便像是在真實世界中作畫一般。但是由於理想運 動軌跡和真實運動軌跡的差異,我們所繪製的光影塗鴉之內容經常因 為過於雜亂而導致無法辨識。在本篇論文之中,我們提供一個藉由樣 本學習的方法來合成長曝光照片之系統去改善上述的問題。我們憑藉 使用者輸入的筆劃和輸入影片來合成長曝光之照片。並且利用影片的 時空線索,我們可以修改光源運動的路徑、線段型式、筆劃粗細,但 卻能夠保持光源與環境間的互動關係。除此之外,我們更可以結合以 上的修改方式合成一段令人驚豔的動畫影片。
Light painting is a photographic technique, where photographer takes a long exposure to record the light motion in the low light environment, like painting the world with light. However, the painting results are always unrecognizable because there are many differences between the exact motion paths we painted and ideal motion paths we wanted. In this thesis, we present a framework for synthesizing realistic long exposure images using example-based method to solve the problem above. In our approach , we synthesize the long exposure image guided by user strokes from input frames. With the spatial-temporal information of light source, we can edit the result of light painting such as its motion, color, line type, and scale while it still remains realistic interaction with environment. Moreover, we can combine the editing techniques mentioned above to synthesize a fabulous animation as its application.