Removal of Impulse Noise from Corrupted Images Using Fuzzy Filter

[[abstract]]摘要 在本篇論文當中，我們提出一個新的模糊濾波器，它擁有移除灰階影像脈衝 雜訊與保護影像細節的能力，與其他方法比較，需要較少的運算時間。為了達到 這些目標，首先我們設計了在受到雜訊干擾的灰階影像中偵測每一個像素是否為 受脈衝雜訊干擾的像素之判斷法則，此判斷法則可減少在模糊濾波器中的運算時 間。在此判斷法則裡，我們計算出3×3 大小視窗裡全部9 個像素與中間鄰近8 個像 素的標準差的絕對差值，絕對差值越小則表示為均勻的像素，反之則為不均勻的 像素且認定為受脈衝雜訊干擾的像素，我們利用此判斷法則可減少掉處理一般像 素的時間。然後，我們提出一個新的模糊濾波器來移除脈衝雜訊與保護影像細節， 它是植基於Farbiz 等人於2000 年所提出的模糊控制濾波器。在新的模糊濾波器當 中，我們在3×3 大小視窗裡8 個鄰近像素的亮度各和中間像素的亮度之絕對差值 裡，取出最小的四個亮度差值做模糊控制濾波器的運算，且在模糊規則庫中使用” 或”運算子，此一作法可以有效移除灰階影像裡大部分的脈衝雜訊與保護影像細 節。從我們的實驗結果顯現出，本篇論文所提出的方法可以達到消除脈衝雜訊與 保護影像細節的目標，且在運算時間需求上比其他方法少。[[abstract]]ABSTRACT In the thesis, we propose a new fuzzy filter which has not only the abilities of reducing impulse noise in gray level image and preserving image details but also less operation time than other methods. To achieve these goals, we first design the decision rule of detecting the impulse noise in the corrupted image. It can reduce the operation time of fuzzy filter with the decision rule. In the decision rule, we first compute the absolute difference of standard deviation of all nine pixels from that of the neighboring eight pixels of central pixel in the 3×3 window. The central pixel in the 3×3 window with low absolute difference is determined as a homogeneous pixel, else it is determined as a non-homogeneous pixel which is treated as a pixel corrupted by impulse noise. Next, we propose a new fuzzy filter to remove impulse noise and preserve image details. It is based on fuzzy-control-filter (FCF) proposed by Farbiz etc. in 2000. In the new fuzzy filter, we take four luminance differences with the smallest iii magnitude from eight luminance differences between neighboring pixels and the central pixel in the 3×3 window to operate with fuzzy-control-filter (FCF), and use “OR” operator in the fuzzy rule-base. By this way, it can effectively remove the impulse noise in the corrupted image and preserve image details. In the experimental results, it is shown that the method proposed in this thesis can achieve the goals of reducing impulse noise, preserving image details, and having less operation time than other methods