臺灣博碩士論文加值系統

本研究是將影像轉換至IHS色彩空間來進行土地覆蓋的分類. IHS色彩空間
為感知式的色彩空間, 它較符合人眼對色彩知覺的判斷, 因此希望藉由將
影像轉換至IHS色彩空間來模擬人類對色彩的判斷.本研究之方法乃是對在
RGB以及IHS中的Raines(1977)和CIE(L*u*v*)三種模型的影像進行分類並
做探討. 使用之遙測影像資料, 一為實驗田圃數化錄影資料, 另一為
SPOT Level 1B 的影像資料. 分類的方法採用監督式的K 均值法. 在精度
評估方面, 以誤差矩陣的整體精度, 標準化誤差矩陣的整體精度及同意係
數的kappa和Tau為指標, 並以Z測試來評估兩獨立同意係數指標之差異性.

In some applications, minpulation color is found to be more
efficient in the perceptual color spaces. The performance of
image classification in the perceptual color spaces is
investigated for remotely sensed images. Both supervised and
non-supervised vonventional classification schemes are applied.
Three bands are selected form the test images and transformed
into IHS color space for classification. The classification
results are then comapred with those obtained with the original
images. From the experiments conducted with two different
images, neither the CIE derived nor the geometry based IHS
color space has demonstrated better accuracy. The results are
analyzed with both kappa and Tau statistics, and Z-test is
performed.