二維經驗模態分解用於影像處理的探討

Abstract

經驗模態分解法(Empirical Mode Decomposition, EMD)的概念已經被採用在希爾伯-黃轉換(Hilbert-Huang Transform, HHT)中將訊號分解成本質模態函數(Intrinsic Mode Function, IMF)，EMD法的優點已在廣泛的研究領域中被報導出來，因此我們嘗試調查二維經驗模態分解法(Bidimensional Empirical Mode Decomposition, BEMD)處理受到高頻雜訊所污染影像的可行性。舉例來說，在我們的研究中，微血管的圖片常受到高頻雜訊的干擾。本論文提出BEMD在影像去雜訊上的初步研究，使用的測試影像包括低細節、中細節以及高細節程度影像，並且加入不同雜訊變異數的高斯雜訊，我們提出兩種影像去雜訊的方法：方法一、徹底移除在影像中所包含主要的高頻成份，IMF1。方法二、藉由門檻值的方法，來移除部份IMF1，方法二主要在保留影像的高頻特性。研究結果也會與適應性濾波器做比較。

Concept and scheme of empirical mode decomposition (EMD) have been adopted to decompose the signal into intrinsic mode functions (IMF) in the HHT (Hilbert-Huang Transform). The advantages of EMD scheme have been reported in a wide scope of research fields. We thus attempted to investigate the feasibility of bidimensional empirical mode decomposition (BEMD) in processing the empirical image data contaminated by high-frequency noise. For example, the capillary imagery in our research of microcirculation system is often interfered by the high-frequency noise. This thesis launched the preliminary study on BEMD employed in image denoising. The tested images include low-detail, medium-detail, and high-detail with additive Gaussian noise with different noise variances. We investigated the performance of two methods in image denoising: 1) (Method I) fully removing IMF1 containing mainly the highest-frequency component in the image, and 2) (Method II) partially removing IMF1 by Thresholding scheme. The design of Method II is aimed to preserve the high-frequency features of the image. The results were also compared with which of the conventional adaptive filtering method.