Abstract
中間值濾波是一種非線性濾波技術, 可以很有效的移除訊號中的雜訊而不 會影響到訊號中劇烈變化的部分, 因此被很廣泛的應用在聲音和影像處理 上。最近, 芬蘭的 Neuvo 首次介紹向量中間值的觀念, 這種方式是將輸 入的向量訊號以向量的方法加以處理, 而不是將每個向量拆成數個不同 的純量再分別處理。向量中間值利用了向量中每一成分之間的相關性, 可 應用在彩色影像, 電視和高畫質電視 (HDTV) 的彩色影像處理上。本論文 中, 我們首先定義 vector median filer, weighted vector median filter, extended vector median filter, vector FIR-median hybrid filter 和三維平面濾波法。三維平面濾波可用在影像序列的處理上, 例 如 HDTV 影像序列, 它將影像序列看成是一個三維的影像, 再用向量中間 值來處理此三維影像。向量中間值的計算非常的浪費時間, 它需要計算每 個不同的向量訊號中兩兩之間的距離, 我們提供一個快速的計算架構來增 快計算速度, 此計算架構可以用硬體或軟體來建立, 其主要的觀念是重覆 利用上一次計算向量中間值之部份結果來達到加速的目的, 並且可用平行 處理的方式來完成。另外, 我們提出幾種近似向量中間值濾波的方法, 它 們可以更有效的移除雜訊或加快移除雜訊的速度。這些方法都試圖避免濾 掉原始訊號中不是雜訊的部分, 以保留訊號中較細微及劇烈變化的部分, 如此可以保持影像品質, 不會使影像變得模糊。 Median filtering is a nonlinear filtering technique which is known for preserving sharp changes in signals and being effective in removing impulse noise. It is very widely used in and image processing since it was first described by Tukey. Neuvo in Finland first introduces the concepts of vector median. this approach, the samples of the vector-value input signal are processed as vectors instead of componentwise scalar processing. vector median utilize the correlation between different This makes it a good approach for HDTV color image processing. In this thesis, we first address vector median filter, vector median filter, extended vector median filter, vector hybrid(VFMH) filter for a color image and 3-dimensional plannar, 3 dimensional multilevel filter for a color image sequence such HDTV image sequence. The computational time of vector filter is very time consuming since it requires to find the minimum in distances bewteen two pixels in the processing window. Thus, a computational architecture is proposed. Also, we also proposed 5 of approximation of vector median filter. They can also smooth very effectively.