消除影像熱像素雜訊之選擇性向量中值濾波器

Abstract

在很多實際用途的光學系統中，多數的商業影像產品，如：數位相機、影像手機和掃描器……等,常發生受到熱像素雜訊干擾的缺陷。而熱像素雜訊的產生是由於CCD光感應器的漏電流現象或影像產品在黑暗或燈光不足的環境下，歷經一段長時間的曝光所導致的。因此，發展熱像素雜訊的移除技術的方法，是必須且重要的。本論文提出消除影像中熱像素雜訊之選擇性向量中值濾波器。觀察大部分的熱像素雜訊，它會在影像中產生是由相鄰的的五到七個偏白的像素所組成的一偏白團。一般的傳統中值濾波器是無法去除這種雜訊的。我們首先提出一套選擇性機制，偵測影像區塊是否為熱像素雜訊所污染，如果影像區塊為熱像素雜訊所污染時，則用我們所提出的向量中值濾波去來處理並取代雜訊區塊；因為有選擇性的取代，所以能避免不需要的處理與取代，然而盡可能的維持原影像的結構與細微處。由實驗結果證明，我們所提出的消除熱像素雜訊之選擇向量中值濾波器能有效移除熱像素雜訊，且能維持濾過影像的細節及結構。在熱像素移除效果比較方面，我們的方法與網際網路網站中的兩個方法，我們處理後的影像在雜訊比(RGB彩色格式)遠勝上述兩種方法。

In many practical image systems, the hot spot like noise occurs in commercial products, such as digital camera, mobile camera, scanner, and so on. The hot spot like noise causes from the imperfection of the sensors of these products or long period exposure in the dark environment. Therefore, it is necessary and important to develop a technique to remove hot spot like noise. The thesis introduces decision-based vectorized median filter for hot pixel like noise removal of images. Most hot pixel like noise is composed of five to seven neighboring pixels forming a white blob. The conventional median filter fails reduce such noise. We first use the decision-based mechanism to detect the image block to check whether the image block is hot pixel noise contaminated or not. If the block is corrupted by the hot pixel noise, the new vectorized median filter is applied to replace the noisy block. The decision-based mechanism will prevent us from unnecessary block replacement and hence retains the original image structure as much as possible. By the simulation results, the proposed decision-based vectorized median filter has demonstrated the effectiveness in the hot pixel noise removal and also pertained finer details and structures well in the filtered image. In comparison to two famous hot pixel noise removal schemes available in the world wide web, our proposed method outperforms these two schemes in PSNR measure (RGB color format) by a large margin.