彩色影像之本質影像研究

Abstract

利用計算最小熵值來找出本質影像的方法，確實可以找到正確的投影角度進而得到本質影像，但是這個方法的執行效率並不好，而且只有某些影像可以找到正確的投影角度，無法針對大部份的影像進行處理。 本篇論文提出以主要成份分析法（PCA）、獨立成份分析法（ICA）與Sobel邊緣檢測法（Sobel Edge Detection）分別來找出正確的投影角度。經過比較之後，利用Sobel邊緣檢測法，從影像中的光線強度變化，可以很有效率地找出正確的投影角度，不只是降低運算量，也能夠針對大部份的影像進行處理。

The algorithm proposed by Dr. G.D. Finlayson computes the entropy to search the angle of the projection for the intrinsic image. After iterative computing, their algorithm indeed gets the correct angle of the projection, which can be used to compute the intrinsic image. Nevertheless, their algorithm spends a lot of time searching the angle of the projection and doesn’t work well for many other images. In this thesis, Principal Component Analysis, Independent Component Analysis, and Sobel Edge Detection are used to search the correct projection angle for the intrinsic image. In the comparison of the results of the proposed three methods, Sobel Edge Detection can efficiently compute the correct angle of the projection by finding the change of the illuminant. Sobel Edge Detection may not only compute quickly with lighter computational load but also work well for a larger set of images.