數位色彩與影像技術應用於視光學之研究

Abstract

在本論文中，我們探討了色彩以及頻譜資訊之間的空間轉換關係。我們利用主成分分析法將高維度的頻譜空間投影並壓縮至三維色彩空間的同時，並保留大部分的資訊量。以虹膜頻譜影像為例，由於人眼無法用於過高的曝光條件，我們捨棄傳統利用空間或時間掃描的機制，利用主成分分析法與最小平方差數位技術將虹膜的色彩資訊經由偽逆矩陣重建其頻譜影像。此外，因為虹膜色彩具有多基因遺傳特徵，我們在本篇論文之中提出利用虹膜的色彩資訊對於頭痛病徵之關聯進行數據分析。我們試圖運用機器學習的方式對於叢發性頭痛患者與正常人進行分類。透過邏輯回歸訓練出來的模型，對於初步收集(27位叢發性頭痛患者)的頭痛資料進行分類，可達70%分類的預測成功率。

In this work, we studied the correspondence between the spectra and color infor-mation. Through the principal component analysis (PCA), we found the high dimensional spectra space can be compressed to the three dimensional color space while keeping most information. For the purpose of optometry, where the strong illumination is inappropriate for the iris imaging. We successfully reconstruct the spectral iridal imaging through the color information by mean of the PCA and least square approximation. In additional to the spectral estimation, this thesis also pioneered the possibility of head-ache classification through the color information. Since the iris color was known as an inherited trait via multiple genes. We tried to use machine learning to distinguish the cluster headache patients from the normal subjects. With preliminary headache data-base, 70% accuracy was achieved to classify the cluster patients and normal person.