臺灣博碩士論文加值系統

MPEG與H.26X、H.32X是目前最主要的動態影像壓縮標準系列，其壓縮編碼的原理可以概分為三個步驟：(1)位移補償、(2)DCT轉換及量化、(3)可變長度類別的Huffman編碼。但還原影像因為量化之程序可能會產生失真及變形，於是我們提出了在前述三個步驟後，再對影像做形態補償，以改善影像品質。形態補償之運算是藉由形態濾波器來達到，因而形態濾波器的種類就是被編碼的對象，所以影像品質改善的資訊是藏在濾波器種類的編碼中。
形態影像處理是利用結構元素去修飾影像的一種有效方法，已成功的應用到電腦視覺、工業檢測上，在本論文是利用其做形態補償的編碼。所以形態補償最主要的就是1.設計快速簡單的形態濾波器與2.最佳形態濾波器的有效選擇。關於前者我們設計了以遮罩為主的濾波器，後者的選擇包含兩個問題，首先就是濾波器候選人的決定，其次就是具有決定權的選民之投票。
對於二階影像，我們利用基底簡化核心解決了第一個問題，又以結構元素用遮罩代替的觀念，解決了第二問題。對於灰階影像，雖然一般權重視窗計算繁難，但我們要因採用遮罩計算簡單的設計，第一個問題如同二階影像一樣解決，第二即選民的問題又以對灰階影像切片的方式來解決。所以我們的遮罩灰階形態濾波器比一般的灰階形態運算方便快速。
最後我們將形態補償的觀念加入實際的MPEG-1軟體系統中並且成功完成播放。實驗證明，形態補償確實能達到改進影像品質之效果，而形態補償之時間問題，我們也以減少濾波器的方式加以改善。

MPEG、H.26X and H.32X are the primary standards for dynamic image compression. There are three stages involved in the MPEG encode process: (1)motion compensation、(2)block DCT transformation with quantization and (3)Huffman coding for run-length categories.The error can be introduced in the quantization process involed in the above second stage. In this sense, a novel idea of shape compensation is proposed by us to adjust the intensity distribution of the motion compensated image. That is, our dynamic images are coded by the kinds of shape transformations in addition to motion vectors after DCT coding.
The shape transformation in this paper is coded by the kinds of morphological operations to be applied. This processing is a type of operation by which the spatial form or structure of objects within an image are modified. Morphological operation is usually applied to the binary images. There are two problems for the selection of the optimal morphological filter: the collection of the candidate filters and the sources of the voters.
For binary images, using the simple structure of the basis of kernel associated to the mask operation and the masked samples in the image set respectively, we solve the candidates and the voters problems. Thus, We have successfully applied the concept of shape compensation to video coding by using the morphological operation. For the gray level images the mask operation is changed to be the more complex window weighting operation. A simplified gray morphological operation by finding the extreme value in the masked neighborhood is proposed for the shape compensation operation. The advantage of mask operation is thus kept for our gray image processing. Therefore, the candidates and the voters'' problem can be solved by a strategy of slicing the image umbrella. Furthermore, our masked gray morphological operation is also more computation-efficient than the regular gray morphological operation.
The shape compensation is integrated into MPEG-1 system in software implementation by us.Experiment results have demonstrated that images going through shape compensation followed by motion compensation have better visual quality than images processed by motion compensation only.Finally we improving speed efficiency by filter reduction to make real-time encoding possible.

目 錄
第一章 引言1
第二章 MPEG動態影像壓縮原理4
第一節MPEG簡介4
第二節MPEG的視訊壓縮原理壓縮原理5
第三章 形態濾波器介紹10
第一節 形態運算介紹11
3.1.1 基本運算定義11
3.1.2 黑白影像之擴張運算11
3.1.3 黑白影像之侵蝕運算12
3.1.4 灰階影像的擴張及侵蝕運算12
第二節 形態濾波器13
3.2.1 遞增式 13
3.2.2 核心之基底16
第三節 建立於基底搜尋的形態濾波器17
3.3.1 形態濾波器之核心17
3.3.2 簡化核心之基底18
第四節 搜尋基底圖簡化核心20
第四章 最佳形態濾波器之決定23
第一節 簡介23
第二節 最佳濾波器之決定23
4.2.1 濾波器之遮罩運算23
4.2.2 利用選民選出最佳濾波器24
第三節 建立機率表25
第四節 利用機率表選出最佳濾波器26
第五節 實際例子說明27
第五章 形態補償之介紹30
第一節 二階形態補償31
第二節 灰階形態補償32
5.2.1 解決灰階問題之做法32
5.2.2 灰階切片之理論基礎36
第三節 減少濾波器的做法38
第六章 系統實作42
第一節 簡介42
第二節 Berkely MPEG-1系統介紹42
第三節 加入形態濾波器45
第四節 MPEG stream之構造47
第五節 解碼端流程47
第七章 實驗結果與討論49
第一節 連續灰階動態影像之形態補償49
第二節 應用形態濾波器於Berkely-MPEG-1系統中51
7.2.1 實驗環境說明51
7.2.2 實驗參數配置53
7.2.3 實驗結果顯示57
第三節 結果討論70
第八章 結論71

[1] Rafael C.Gonzalez and Richard E.Woods, "Morphology",Digital Image Processing,Copyright 1992 by Addision-Wesley Publishing Company ,Inc,p518-p548.
[2] Jain,A.K.Fundamentals of Digital Image Processing,Prentice-Hall,Englewood Cliffs,N,J,1989
.
[3] Sperry R.H.and Parker K.J. "Segementation of speckle images based on levelcrossing statistics"J.Opt.Soc.AM.A/vol.8,No.3/March 1991,pp.490-498.
[4] Klinger J.W.,Vaughan,C.L.and Andrews L.T. "Segmentation of echocardiographic images using mathematical morphology",IEEE Trans.on Biomedical engineering,vol.35,NO.11,NOV.1988,pp925-933
[5] Charles R.Giardina and Edward R.Doungerty, "Morphology Filters for Two-Valued Images", Morphological Methods in Image and Signal Processing,Prentice Hall,Englewood Cliffs,New Jersy 07632.p133-p155.
[6] Charles R.Giardina and Edward R.Doungerty, " Increasing ", Morphological Methods in Image and Signal Processing,Prentice Hall,Englewood Cliffs,New Jersy 07632.p133-p136.
[7] Charles R.Giardina and Edward R.Doungerty, " Appendix:Proposition A.39", Morphological Methods in Image and Signal Processing,Prentice Hall,Englewood Cliffs,New Jersy 07632.p13.
[8] Charles R.Giardina and Edward R.Doungerty, " Basis for Kernel", Morphological Methods in Image and Signal Processing,Prentice Hall,Englewood Cliffs,New Jersy 07632.p142-146.
[9] E.R.Doughert and C.R.Giardina,"A Digital Version of the Matheron Representation theorem for Increasing in Terms of a Basis for the Kernerl",IEEE Computer Vision and Pattern Recognition(1986),p534-536.
[10]Edward R.Doungerty, "Optimal Mean-Squre N-Obsevation Digital Morphological Filters,I, Optimal Binary Filters",in Computer Vision and Graph Image Processing,vol.55,No.1.January, pp36-54,1992.
[11]Charles R.Giardina and Edward R.Doungerty, "Umbra transform", Morphological Methods in Image and Signal Processing,Prentice Hall,Englewood Cliffs,New Jersy 07632.p174-181.
[12]Bede Liu and Andre'' Zaccarin, "New Fast Algorithms for the Estimation of Block Motion Vectors," IEEE Trans. on Circuits and Systems for Video Technology,Vol.3,No.2,pp.148-157 April,1993.
[13]Ram Srinivasan and K. R. Rao, "Rredictive Coding Based on Efficient Motion Estimation," IEEE Trans. On Commun., Vol.COM-33,No.8,pp.888-896, Aug.1985.
[14]S.Kappagantula and K. R. Rao,"Motion Compensated Interframe Image Predicition," IEEE Trans. On Commun.,Vol.COM-33,No.9,pp. 1011-1015, Sep.1985.