本篇論文針對傳統的經驗模態分解法 (Empirical Mode Decomposition, EMD) 提出一個改良的方法。不同於傳統的經驗模態分解法，本研究所提出的方法可節省大量的計算時間，將其應用在光影不均的影像強化時，幾乎可達到傳統的經驗模態分解法所能達到的效果。 本研究並將此改良的方法應用在目前較少有相關文獻探討到的問題─背影透光的文件影像，這是其它光影校正的方法所無法解決的問題，例如，直方圖均化 (Histogram Equalization)。 經實驗結果證明，本研究所提出的方法在文件影像受光的影響下，其所可能產生的問題，例如，曝光不足、曝光過度、不均勻光場、背影透光等問題均可有效的解決。 In this thesis, an improved technique of Empirical Mode Decomposition(EMD) is proposed. Different from traditional empirical mode decomposition, the computational time of the improved technique can be greatly reduced to ensure real time processing. Moreover, the improved technique almost possesses the same result as the traditional empirical mode decomposition when applying the technique in removing uneven illumination for image enhancement with application in 3D OCR. We also employ the improved technique to remove background texts of document images caused by illumination, which was seldom discussed in previous literatures. It is still an unresolved problem by using other methods such as histogram equalization. As can be seen in the experimental results, the improved technique works well for document images in various illumination environments, such as overall darkness, overall brightness, uneven illumination, and background text document image. This work is a prerequisite to facilitate and ensure the success of later 3D OCR.
