位移資訊在動態影像分析中是很重要的參數。利用光流計算法來求取位移，以廣泛地應用到計算機視覺的領域中。本論文利用光流求得兩張圖片間的位移資訊，使用慣性作用的假設,希望藉由第一、二張圖片的位移資訊預測第二張和第三張圖片的位移資訊，利用預估位移資訊可求得預估圖。利用此張預估圖當下一張編碼圖片的參考圖。當預估圖與下一張原圖差異減少時,便可減少補償資訊的紀錄。
我們改良HORN&SCHUNCK所提出的光流法，將分割的特性加入光流演算法中，讓求得的位移資訊更具群體性，可簡化資料上的紀錄。改良式光流在還原圖片上比傳統光流更理想，且因為圖片的預估是假設物體均以等速移動，但動態影像描述的移動物體並不全然符合這個假設。於產生第三張圖片預估圖方向上，本實驗提出兩種方法做比較,其一全然採用第一、二張的速度資訊。其二採用第一二張所求的速度資訊形狀,但卻要更新區域的速度值。實驗結果顯示我們的方法在壓縮率上比傳統動態補償有1%的改進。

Displacement information is important in dynamic image analysis. The method of optical flow has been well applied to compute the displacement in the field of computer vision. We apply the method of optical flow to compute the displacement information for video compression. We can predict the optical flow between picture 2 and picture 3 by the optical flow between picture 1 and picture 2 by using the principle of inertia. Using the predicted optical flow between picture 2 and picture 3, we can recover a rough version of picture 3. This version can be taken as a reference picture for encoding picture 3. This reference will decrease the compensation information in the following stage, and then improve the compression ratio of MPEG .

We modified the traditional optical flow of Horn & Schunck to a regional optical
flow by segmentation. Then, the displacement information could be reduced. The
picture is recovered by the optical flow in a modified way because some objects
couldn’t move in the same direction and velocity. We propose two methods in
optical flow prediction. One is the complete information: shape and value. The
other is shape only with value recomputed and extra encoded. Experiments
demonstrate a better compression ratio of 1% for our motion compensation than
the regular motion compensation.

目 錄

第一章 引言 ．．．．．．．．．．．．．．．．．．．．．．．．．． 1

第二章 MPEG動態影像壓縮的原理 ．．．．．．．．．．． 6
2-1 MPEG簡介 ．．．．．．．．．．．．．．．．．．．．． 6
2-2 MPEG的視訊壓縮原理 ．．．．．．．．．．．．．．． 7

第三章 光流與分割之基本定理 ．．．．．．．．．．．．．．．．．．．． 12
3-1 簡介 ．．．．．．．．．．．．．．．．．．．．．．．．．． 12
3-2光流之基本定理 – 傳統方法 ．．．．．．．．．．．．．．．． 12
3-3 區域成長之重心聯合基本定理 ．．．．．．．．．．．．．．．． 22

第四章 分割式光流法原理介紹 ．．．．．．．．．．．．．．． 26
4-1 簡介 ．．．．．．．．．．．．．．．．．．．．．．．．． 26
4-2 分割式光流法的基本定理 ．．．．．．．．．．．．． 26
4-3 光流的梯度求法 ．．．．．．．．．．．．．．．．．．． 34

第五章 系統實作 ．．．．．．．．．．．．．．．．．．．．．．． 37
5-1 簡介 ．．．．．．．．．．．．．．．．．．．．．．． 37
5-2 分割式光流法的計算 ．．．．．．．．．．．．．． 37
5-3 動態影像編碼流程 ．．．．．．．．．．．．．．． 46
5-4 動態影像解碼流程 ．．．．．．．．．．．．．．．． 47

第六章 實驗結果與分析 ．．．．．．．．．．．．．．．．．．．． 48
6-1 簡介 ．．．．．．．．．．．．．．．．．．．．．．．．．． 48
6-2分割式光流和傳統光流在圖片還原上的比較 ．．．．．． 49
6-3 預測圖片之比較 ．．．．．．．．．．．．．．．．．．．． 52
6-4壓縮率和訊雜比之實驗結果與討論 ．．．．．．．． 61

第七章 結論 ．．．．．．．．．．．．．．．．．．．．．．．．．． 67

[1].Motioncompensated video compression using adaptive ransformations Diab, Z.;
Cohen, P. Acoustics, Speech, and Signal Processing, 1997. ICASSP-97., 1997
IEEE International Conference on , Volume: 4 , 1997 Page(s): 2881 -2884
vol.4
[2].Region growing motion segmentation and estimation in object-oriented video
coding Yemez, Y.; Sankur, B.; Anarim, E. Image Processing, 1996.
Proceedings., International Conference on , Volume: 1 , 1996 Page(s): 521 -
524 vol.2
[3].An optical flow based motion compensation algorithm for very low bit-rate
video coding Shu Lin; Shi, Y.Q.; Ya-Qin Zhang Acoustics, Speech, and Signal
Processing, 1997. ICASSP-97., 1997 IEEE International Conference on ,
Volume: 4 , 1997 Page(s): 2869 -2872 vol.4
[4].A low bit-rate video codec based on two-dimensional mesh motion
compensation with adaptive interpolation Pohsiang Hsu; Liu, K.J.R.; Chen,
T. Circuits and Systems for Video Technology, IEEE Transactions on ,
Volume: 11 Issue: 1 , Jan. 2001 Page(s): 111 -117
[5].2-D motion estimation with hierarchical content based meshes
Al-Regib, G.; Altunbasak, Y. Acoustics, Speech, and Signal Processing,
2001. Proceedings. 2001 IEEE International Conference on , Volume: 3 , 2001
Page(s): 1621 -1624 vol.3
[6].B . G . Schunck，〝Image flow : Fundamentals and algorithm〞，in motion
Understanding : Robot and Human Vision ， W . N .Martin and J . K .
Aggarwal ，Eds . Norwell ，MA : Kluwer Academic Publishers ，1988 。
[7].J . K . Aggarwel ，L .S .Davis ， and W. N. Martin ，〝Correspondence
processes in dynamic scence analysis，〞 Proc.IEEE，vol.69，no.6，pp.562-
572，May 1981。
[8].J. Q. Fang and T. s. Huang ，〝Some exeperiments on estimating the 3-D
motion parameters of a rigid body from two consecutive image frames，〞IEEE
Trans. Pattern Anal。 Machine Intell. Vol.PAMI-6 ，pp.545-554，Sept.1984。
[9].S. Ranade and A. Rosenfeld ， 〝Point pattern matching by relaxation，〞
Pattern Recognition ，vol. 12 ，pp. 269-275，1980。
[10]. R.C.Dubes, A.K.Jain, ”Random Field Models in Image analysis,”
J.Applied Statistics, Vol.16,No.2,pp.131-164,1989。
[11].Ben-Shung Chow, Seh-Shion Huang, and Yuan-Chuan Chang, ” Structure
of Computing Image Texture Function and Its Methods,” HD-
Media'94,Taipei,1994。
[12].Louis L. Scharf,Statistical Signal Processing: detection ,estimation, and
time series analysis, Addison -Wesley Pub. Co.,1991。
[13].Gibson， J ，J.，on the analysis of change in the optic array ，
Scandinavian J.Psychol.18(1977)161-163。
[14].Nakayama，K.and Loomis，J.M.，Optical velocity patterns ， velocity-
sensitive neurons and space perception，Perceptrion3(1974) 63-80。
[15].T.Pavlidis, Structure Pattern Recognition, Springer, New York , 1977。
[16].R.M.Haralick and Linda. G.Shapiro, Computer and Robot Vision ,
Vol.1,Addison-Wesley Pub.Co.,1992。
[17].S.L.Horowitz and Y.Pavlidis, ”Picture Segmentation by a Directed Split-
and-Merge Procedure,” Proc. 2nd Int. Joint Conf. Pattern
Recognition,pp.424-433,1974。
[18].S.W.Zucker, ”Region Growing : Childhood and adolescence,” Computer
Graphics and Image Processing,5,pp.382-399, 1976。