基於邊緣偵測之短波紅外線影像壞點偵測與校正

Abstract

本論文使用邊緣偵測和模糊規則偵測短波紅外線影像壞點像素並藉由兩次中值濾波器校正壞點像素。此外，對於短波紅外線影像像素非均勻現象，我們使用兩點式校正方法。 為了提高壞點校正效能，我們也提出改進基於向量次序統計之彩色邊緣偵測技術的方法，我們的邊緣偵測方法包含兩個部份，第一部份，我們利用模糊梯度的概念來估測每個處理像素的梯度方向，並且根據此方向來調整相對應的視窗方位；第二部部分依向量次序統計計算向量平均距離(VMD)，如此一來，整合了向量次序統計與模糊梯度的偵測方法，因此能夠產生更為穩健的邊緣偵測響應。更進一步，我們將此技術整合到一個新的門檻偵測方法，此方法依據影像內容自動作最佳化調整門檻，而不需要手動選取。由測試彩色合成影像與實際影像的數據顯示，我們的自動彩色邊緣偵測是非常方便與可靠的，期許能夠更進一步提高紅外線影像壞點偵測效能。

In this thesis, we first use edge detection and fuzzy rules to find bad pixel map of a SWIR sensor. Then we employ two median filters sequentially to correct them. Moreover, we apply two-point correction method to correct non-uniformity among pixels of SWIR sensor. To enhance the tools for bad pixel correction, we have also proposed a new color edge detector based on vector order statistics. The proposed detector consists of two stages. In the first stage, we use fuzzy gradient to estimate the direction of the gradient for every processing pixel in the image and adjust the corresponding processing window according to this detected direction for reliable edge detection setup. The second stage computes the vector mean distance (VMD) based on vector order statistics. Hence, the proposed detector, which integrates vector order statistics and fuzzy gradient, can provide more robust response for edge detection. Furthermore, we also combine the edge detector to our proposed thresholding method, which can automatically determine an optimal threshold and be adaptive to different image contents without manual intervention. Thus, the excellent results by our proposed edge detection scheme demonstrate that it is very user friendly and confident. This edge detection scheme could also be promising for better detecting bad pixels of a SWIR image sensor.