利用色彩資訊之影像切割方法研究

Abstract

目前影像切割的技術大多著重在灰階影像，而我們認為利用色彩資訊能得到更多的資訊來幫助我們切割影像。在這篇論文中，我們討論了對色彩空間的選擇，也討論了目前影像切割技術中可能存在的問題。根據這些討論，我們設計了一套新的影像切割方法。我們先將影像中在三維色彩空間的色彩資訊轉換成在亮度資訊和在二維色彩空間的色彩資訊，然後利用我們所提出的天秤運算子(Balance Operator)來找出影像中趨勢的變化，再根據這些趨勢先把影像中對比值資料較小的區域進行第一次合併，而對比值較大的地方則以分配(Allotment)的方式來處理，最後再進行第二次合併。經由一些初步的實驗結果顯示，這套新的影像切割方法似乎能夠比較正確地進行彩色影像的切割。

So far, most of image segmentation algorithms are focusing on the segmentation of gray-level images. However, more information could be gotten if we employ color information in segmenting images. In this thesis, we discuss the selection of color space first and then discuss the problems which occur in current image segmentation algorithms. According to these discussions, we design a new image segmentation algorithm. In this algorithm, we convert the color information of the image from the original 3-D RGB color space to the (I, R-G, Y-B) color space. Then, we propose the Balance Operator to detect the trend of information in the image. According to these trends, we merge low contrast regions first and process high contrast regions by using the allotment method. Finally, we merge together the regions that have similar intensity and color. Base on some simulation results, it seems that we can more correctly segment images by using this color image segmentation algorithm. Chapter 2 Backgrounds 2.1 Previous Algorithms for Color Image Segmentation 2.2 Human’s Color Vision Chapter 3 Color Space and Balance Operator 3.1 Color Space 3.2 Balance Operator 3.2.1 Balance Operator vs. 2nd-Order Differentiation 3.2.2 Zero-Crossing and Local Extreme Chapter 4 Image Segmentation Algorithm 4.1 Edge Band 4.1.1 Traditional Methods 4.1.2 Edge 4.1.3 Edge Band 4.2 Allotment Method 4.3 2nd-passing Merging Chapter 5 Simulation Chapter 6 Conclusions and Future Work