數位視訊抽圖技術之應用與實作

Application and Implementation of Video Matting Technique

張嘉祐(Chia-Yu Chang)

指導教授 ： 洪國龍

朝陽科技大學/資訊學院/資訊管理系/碩士(2014年)

摘要

數位視訊是指以數位訊息記錄的視訊資料。視訊抽圖技術主要應用於電影的後製及影像的合成，在以往皆需要大量的人力來操作，以及昂貴的成本付出，對於將前景物件的抽取，尤其是在邊緣的部分，若是要將毛髮或是複雜的邊緣處理的完美，是有相當的難度，這個問題在“A Closed-Form Solution to Natural Image Matting” Levin 2008[1]之論文已詳細討論，本技術報告則為此論文的系統實作，我們利用該論文的技術設計了一套操作軟體，以便於執行視訊抽圖時能更加的簡易及便利。
“A Closed-Form Solution to Natural Image Matting”當中所使用的技術，是以Closed-Form Matting為基礎架構來估計每個像素點的α值，並且將這些演算法延伸使用在視訊抽圖技術上，實驗證明，它可以精確的將前景物件取出。乃如想要達到事半功倍，又能完美呈現抽圖技術的效果，有一套好的操作工具是必然的，如我們之成果展示，本技術報告所完成的系統，不僅能夠使用在不同影像上，更能夠精確地估計出視訊當中，每張畫面的alpha matte，並且在合成後的影像或是影片都可以達到一定的品質。

關鍵字

視訊抽圖 ； α值

並列摘要

Digital video is messages that digitally recorded video data. Video matting technique is mainly used for post-production of movie film and image synthesis. In the past, a lot of manpower is needed so that the cost is expensive for the extraction of the foreground object, especially in the edge portion. If you want to process the hair or complex edge area perfectly, it will have a considerable difficulty. This issue had been discussed in detail in the paper titled “A Closed-Form Solution to Natural Image Matting”, Levin 2008 [1]. As for this technical report, it is the implementation of the Levin’s paper. We explore techniques of the paper to design an operating software that it can be able to perform video matting simply and conveniently.
In the paper of “A Closed-Form Solution to Natural Image Matting”, the method based on Closed-Form Matting framework is used to estimate α value for each pixel, and then it is extended to video matting. In the experiment, it had been proved that the method can accurately remove foreground objects effectively. However, if you not only want to achieve efficiency but also to show the effect of the technique perfectly, to have a good operating tool is inevitable. As our results show, the system that technical report implemented not only can be used on different images, but also it can accurately estimate the alpha matte of the frames of the video. Moreover, the synthesized images or videos can reach a certain quality.

並列關鍵字

Video Matting ； Alpha Matte

參考文獻

[1] Levin, A., Lischinski, D., and Weiss, Y.( 2008 ),“A Closed-Form Solution to Natural Image Matting,”Journal of IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.30,no.2,pp.228-242.

[4] Bai, X., and Sapiro, G.( 2007 ), “A Geodesic Framework for Fast Interactive Image,”conference of IEEE Computer Vision, PP.1-8.

[6] Kappagantula, S., and Rao, K.( 1985 ). Motion Compensated Interframe Image Prediction. IEEE Transactions on Communications, vol.33,no.9,pp.1011-1015.

[9] Ruzon, M. A., and Tomasi, C.( 2000), “Alpha estimation in natural images”.In CVPR2000,pp.18-25.

[11] Sarim, M., Hilton, A., Guillemaut, J. Y., Kim, H., and Takai, T.( 2010 ), “Multiple view wide-baseline trimap propagation for natural video matting,”conference of IEEE Visual Media Production,pp.1-8.