快速門檻值選擇方法於多峰影像分塊技術之研究

Study on Multimodal Image Segmentation Technique by Using Fast Thresholding Selection Procedures

李興緯(Hsing-Wei Lee) ； 張奇(Ji Chang) ； 李翔祖(Hsiang-Tsu Li)

《中華印刷科技年報》 2010 (2010/03) Pp. 299-304

https://doi.org/10.30153/JCAGST.201003.0021

摘要

本研究主要在於探討多峰影像分塊方法，從本質上來說，影像分塊是將性質相近之像元進行群聚的過程，分塊所依據的是像元的灰階值(gray-level value)，基於此特性以通過門檻值(threshold)的方式進行分塊；利用影像分塊的技術，可從影像中分離出所需要的組成部分，並可利用此方法從複雜的背景中萃取出所想要的影像特徵。
關於選取門檻值的方法，本研究是以影像的灰階直方圖為依據，假設各群體的灰階在直方圖中呈常態分配，且使用高斯濾波(Gaussian Filter)平滑灰階直方圖，並選取波谷所對應的灰階值為門檻值來進行分塊；若灰階直方圖中的灰階群過份重疊，可能導致平滑後的直方圖呈現單峰的形式，此時則選取單峰直方圖上具有最大曲率的灰階作為門檻值；在去除雜訊方面，則是以兩組門檻值，包括原始影像及平滑影像所求出的門檻值，對影像中任一像元的原始灰階及其鄰域局部平均灰階進行分塊；此外快速門檻值的選取方式將與高斯平滑化法、類別間變異數法及熵法進行比較，評估分塊結果的優劣性。

關鍵字

多峰影像 ； 門檻值 ； 直方圖 ； 高斯濾波 ； 分塊

並列摘要

This research mainly studied on multimodal image segmentation technique. To say about the essence of the image segmentation is a process of clustering the close pixels. It depends on the gray-level values of the pixels to segment by passing the threshold. By using the image segmentation technique, we can not only get the main need parts of the image, but also the characteristics from the complex background of image. About the method of the choosing threshold, the study is based on gray-level histograms. It supposes all of the gray-level values that distributed normality. It uses Gaussian filter to smooth gray-level histograms and chooses the valleys to segment. If the gray-level groups of the histograms overlap too much, the smoothed gray-level histograms may be a single peak. We get the maximum curvature of the single peak histogram to be the threshold. In the part of removing noises, we use two thresholds that included original and smooth images to segment any pixel of original gray-level and the average gray-level of its nearly area. In addition, our method would compare with Gaussian Kernel Smoothing, between-class variance and entropy methods.

並列關鍵字

multimodal image ； threshold ； histogram ； Gaussian filter ； segmentation

參考文獻

Agus,Z.A.,Akira, A.(2006).Image segmentation by histogram thresholding using hierarchical cluster analysis.Pattern Recognition Letters.27(13),1515-1521.

Chen C.J.,Chang R.F.,Moon W.K.,Chen D.R.,Wu H.K.(2006).2-D ultrasound strain images for breast cancer diagnosis using nonrigid subregion registration.Ultrasound in Medicine and Biology.32(6),837-846.

France.Weszka,J.,Rosenfeld, A.,IEEE(Trans.)(1979).Histogram modifications for threshold selection..Syst. Mun Cybernet..9,38-52.

Glasbey,C.A(1993).An analysis of histogram-based thresholding algorithms..CVGIP: Graphical Models and Image Processing.55,532-537.

Hou,Z.,Hu,Q.,Nowinski,W.L.(2006).On minimum variance thresholding.Pattern Recognition Letters.27(14),1732-1743.