進階搜尋


 
系統識別號 U0026-0812200910355885
論文名稱(中文) 以馬可夫模型為基礎之彩色影像分割
論文名稱(英文) Color Image Segmentation Based on Markov Random Field
校院名稱 成功大學
系所名稱(中) 統計學系碩博士班
系所名稱(英) Department of Statistics
學年度 91
學期 2
出版年 92
研究生(中文) 陳俐穎
研究生(英文) Li-Ying Chen
學號 r2690106
學位類別 碩士
語文別 英文
論文頁數 51頁
口試委員 口試委員-李隆安
指導教授-孫永年
指導教授-吳鐵肩
中文關鍵字 彩色影像分割 
英文關鍵字 clustering analysis  Makrov Random Field 
學科別分類
中文摘要 在計算機視覺及影像處理的領域中,影像分割佔了一個很重要的地位。影像分割的目的,是希望能夠找出影像中,我們所感興趣的區域,或者是有意義的區域。彩色影像由三原色所構成,可以視為是三維度變量的資料,而群集分析是一種用來處理多變量資料的分析方法,應用於彩色影像成為一種以區域為基礎的分割方法。其利用量測資料點之相似程度,將像素點區分成數個群集。每一個群集內的資料有較高的相似度,而對應在圖形上,則可以清楚得到某些物體、區域的形狀。在做群集分析之前,必須先設定好其中的兩個參數,一是分群集的數目,另一則為起始群集重心的選定。由於分割之前,我們並沒有影像的資訊 ( 影像中有多少個區域、物體),故我們設計一些準則,由電腦自動決定分割的數目以及起始分割的中心座標。
做完群集分析得到的分割圖像,其中可能含有一些雜訊,或是被誤判為邊界的點,以及沒有被偵測出的邊界點,這部分仍須做修正。已知馬可夫隨機場模型(MRF),可用來處理較低影像品質的圖片,因此我們採用MRF模式來做修正。而在MRF模式中的變數,並不是直接利用彩色圖片的原始資料,而是利用修改過的Sobel運算法得到的結果當成我們的變數,其中並將每一個色彩平面的變異考慮進來。由此運算法得到的數值,較高的數值代表的是有較高的可能性為邊界。
我們利用不同的彩色空間座標,比較得到分割效果的差異。其中比較的準則,是我們自行定義的「偵測一致比率」。我們共討論四種彩色座標平面,包括RGB、OHTA、YIQ、CIELUV。而在我們所提出的方法中,利用RGB座標所得到分割效果最佳。而利用CIELUV 座標所得到分割效果,有較大的變異,較適用於色彩有顯著的差異,或是物體較多的影像。利用本文所提出的方法,在RGB座標平面上,針對一張中等大小的影像,平均約花費6-10分鐘即可。
英文摘要 Image Segmentation is an important field in computer vision and image processing. The target of image segmentation is to define regions that represent the interesting objects or meaningful parts of objects. Color image can be regarded as the three-dimensional data. Clustering analysis is commonly used for analyzing multivariate data and is adapted to color image as a region-based segmentation method. It divides the set of processed patterns into clusters based on the similarity of subset pixels. Each cluster contains patterns representing objects that are similar in the selected descriptions or criteria. There are two key terms to be defined before clustering: the number of clusters and the cluster centroids. Since we don’t have any prior knowledge of the image, we propose two rules to compute the two terms automatically.
After the clustering analysis, some noise or edges may present in locations where there is no boundary, and no edge presents where a real boundary exists. Markov Random Fields (MRF) image model is adopted in processing the degraded image and correct the results from clustering analysis. In MRF, we do not use the intensity value of each color channel directly. Instead, we incorporate the value of revised Sobel operation and the variance within each color space into the energy computation. Therefore the higher value represents high probability to be defined as an edge pixel.
The experimental results are compared under different color space. We defined a criterion called the “consistent detection rate” to evaluate the segmentation results. Four color spaces, RGB, YIQ, CIELUV, and OHTA, are adopted in this study. From the experiments, we conclude that the RGB color space performs well in most of the tested images. CIELUV space is good for the images having significant changes in color or with many objects. It is not recommended to use it in regular images. The proposed method can achieve appropriate segmentation in RGB color space in about six to ten minutes for medium size color images.
論文目次 Chapter 1 Introduction………………………………………1
1.1 Motivation………………………………………………1
1.2 Previous Works…………………………………………1
1.3 Outlines…………………………………………………3
Chapter 2 Color Segmentation………………………………4
2.1 Segmentation…………………………………………4
2.2 Color segmentation…………………………………5
2.3 Color space……………………………………………6
Chapter 3 Regional Clustering………………………………9
3.1 Clustering……………………………………………11
3.1.1 Number of Clusters………………………………12
3.1.2 Initial Seeds………………………………………15
3.2 Revised Sobel Operator……………………………17
3.3 Markov Random Field…………………………………19
3.4 Continuity checking…………………………………20
3.5 Consistency Ratio……………………………………27
Chapter 4 Experimental Results………………………………28
4.1 Experiments with Real Color Data………………28
4.2 Compare with Different Color Space……………35
4.3 Compare with DDMCMC………………………………37
Chapter 5 Conclusion and Future Work……………………48 References………………………………………………………50
參考文獻 [1] J. Besag,” Spatial Interaction and the Statistical Analysis of Lattice Systems,” (with discussions). Journal of the Royal Statistical Society, Series B Vol.36, pp192-236, 1974.
[2] Y. Yakimovsky, “ Boundary and object detection in real world images,” J.ACM, Vol. 23, pp599-618, 1976.
[3]W. E. Larimore, “Statistical interaction and statistical analysis of lattice systems, “Proc IEEE, Vol. 65, [[961-970, 1977.
[4] D. Cooper, “ Stochastic boundary estimation and object recognition,” Computer Graphics and Image Processing, pp326-355, April 1980.
[5]Y. Ohta, T. Kanade, and T. Sakai, “Color information for region segmentation”, Computer Graphics and Image Processing, Vol. 13, pp222-241, 1980.
[6] J. Besag ,”On the statistical analysis of dirty pictures,”(with discussions). Journal of the Royal Statistical Society, Series B, pp259-302, 1986.
[7]J. Silverman and D. Cooper, “Bayesian clustering for unsupervised estimation of surface and texture models, “IEEE Trans. Pattern Analysis and Machine Intelligence”, Vol. 10, No. 4, pp482-495, July 1988.
[8]R. Chellappa, S. Chatterjee, and R. Bagdazian, “Texture synthesis and compression using Gaussian-Markov random field models, “ IEEE Trans. Systems, Man, and Cybernetics”, Vol. 8, No.2, pp298-303, March 1995.
[9] D.K. Panjwani and G..Healey,” Markov Random Field Models for Unsupervised Segmentation of Textured Color Images,” IEEE Trans. on PAMI, Vol. 17, No. 10,pp939-954, Oct 1995.

[10] Z. Tu and S.C. Zhu,” Image Segmentation by Data-Driven Markov Chain Monte Carlo,” IEEE Trans. On PAMI Vol. 24, No. 5,pp657-673, May 2002.
[11] J. Mukherjee,” MRF Clustering for Segmentation of Color Images,” Pattern Recognition Letters 23,pp917-929, June 2002.
[12] S. C. Zhu and X. Liu, “Learning in Gibbsian Fields: How Accurate and How Fast Can It Be?,” IEEE Trans. on PAMI. Vol. 24, No. 7,pp1001-1006, July 2000.
論文全文使用權限
  • 同意授權校內瀏覽/列印電子全文服務,於2004-06-17起公開。
  • 同意授權校外瀏覽/列印電子全文服務,於2006-06-17起公開。


  • 如您有疑問,請聯絡圖書館
    聯絡電話:(06)2757575#65773
    聯絡E-mail:etds@email.ncku.edu.tw