應用模糊分類法於光學衛星影像之對比增揚

Image Enhancement of Optical Satellite Images Using Fuzzy C-Mean Classification

張立雨(L. Y. Chang) ； 陳繼藩(C. F. Chen) ； 張宏宇(H. Y. Chang)

《航測及遙測學刊》 15卷2期 (2010/06) Pp. 155-166

https://doi.org/10.6574/JPRS.2010.15(2).2

摘要

一般用於光學衛星影像的對比增揚方法皆直接使用整張影像的統計資訊對所有灰階值進行增揚處理。但由於傳統的影像增揚方法必須犧牲原影像中較暗與較亮的區域或是出現機率較低之灰值來進行增揚，因此許多不同的地物特徵或細節通常難以同時增顯出來。本研究提出一種以模糊理論爲基礎的影像對比增揚方法，先將單一像元視爲數種類別的組合，並且針對各類別分別進行增揚對應函數之計算，藉以補償上述傳統方法中以全幅影像進行增揚所造成之問題。本方法主要分爲三個階段：第一階段，以Fuzzy C-Means(FCM)群聚分類法對衛星影像作模糊式分類，將原始影像由灰階值空間轉換至歸屬值空間，分類後的各個像元由數個相應於類別比例的歸屬值所組成。第二階段，依照各類別的歸屬值，分別建立各類別的增揚轉換模型。第三階段，將歸屬值依照前一階段中建立的轉換模型轉換回灰階值空間。由於每個像元皆由數個類別的歸屬值組成，原始灰階值會依照各類別的轉換模型被增揚成數個不同的值。因歸屬值代表類別混合的比例，故以各類別歸屬值作爲權重，重新組合此些不同的灰階值，得到最後的增揚成果。影像經增揚後，評估採用定性及定量方式，分別以人眼及量化指標判定增揚影像含有的資訊量及對比度，並將模糊分類式增揚法的成果與傳統常用的非線性直方圖等化及線性對比擴展法的成果比較。成果顯示對光學衛星影像而言，本研究提出的模糊分類對比增揚演算法對視覺而言可以提供較佳的影像品質，且表現比傳統方法更佳。

關鍵字

對比增揚 ； 模糊集合理論 ； Fuzzy C-Means

並列摘要

Many image enhancement algorithms have been developed to improve the appearance of optical satellite images. However, it is usually difficult to enhance all land cover classes appearing in the images, because local contrast information and details may be lost in the dark and bright areas. In order to compensate the local brightness lost in the dark and bright areas, a fuzzy-based image enhancement method is developed to enhance the source image according to its fuzzified class information. The algorithm contains three stages: First, the satellite image is transformed from gray-level space to membership space by Fuzzy C-Means clustering. Second, appropriate stretch model of each cluster is constructed based on corresponding memberships. Third, the image is transformed back to the gray-level space by merging stretched gray values of each cluster. Various satellite images are used to test the proposed algorithm. Since the gray values of some classes are extremely dark or bright, apparently the global enhancement will result in poor contrast quality. After using the proposed enhancement method, the results are evaluated and compared with other conventional methods. The test results indicate that the proposed enhancement method can provide superior appearance and quality than other conventional methods.

並列關鍵字

Contrast enhancement ； Fuzzy set theory ； Fuzzy C-Means