摘要
本論文將探討中值濾波器之電路實現，中值濾波器主要功能在過濾除去信號雜訊，方法簡單效果不錯，其中最主要之電路為處理數列值之大小排序。一般的排序法之電路為規律之網絡架構實現，而本論文採用排序速度較快的插入排序法，將設計其電路架構，並且在過程中，開發出硬體模組化，可適用於不同排序數目。此插入排序法的電路架構設計，使得排序所需的時脈縮減為N筆資料只需要N個時脈即可完成排序；再加上模組化的設計，使操作者需排序N筆資料時，將N個排序單元模組串列組合，即可合成N筆資料的硬體排序電路，排序資料的數量可以為任意數並不侷限於二的冪次方數。對於中值濾波器這類需要應用到9筆資料的排序，此完成的排序模組，可以有效處理奇數筆資料的排序。 本論文將提出的插入排序法電路實現在中值濾波器的設計，利用DE2-115 FPGA開發平台進行系統驗證。透過界面，有雜訊的圖片儲存至開發板的靜態隨機存取記憶體內，除了周邊的像素外，圖片每個像素都必須經中值濾波器處理，讀出像素及其周圍8個像素資料同時輸入至中值濾波器，用以排序此9筆像素資料，且將排序完成的中間數值像素，取代原本像素值。本論文將顯示比較處理雜訊前與完成雜訊後圖像之差異。
並列摘要
We study the circuit implementation of the median filter in this dissertation. The main function of the median filter is to filter the salt-pepper noise. The filter circuit is designed to sort the number list of the mask pixels and to find the middle value which replacing the target pixel. We use the insertion sort method to sort the number list. This method is simple and efficient. The method also can reduce the salt pepper noise in the image. We develop the hardware module to adapt different number data to sort. For the modulization, we change easily to sort different number of data in the hardware circuit. It only needs the N system clocks to sort if we have N number of data. N is not limited by power of 2 or even number. Our proposed the hardware architecture of the insertion sort reduces the sorting time. Our proposed insertion sort circuit also extends easily to different number of list. Thus, the circuit is used to the median filter. The main function of the median filter is to sort 9 pixels of data list. We use the DE2-115 FPGA development board to verify the proposed circuit of the median filter. In the system verification, we store the noise image to the SDRAM on the board. We store the developed HDL code of the median filter into the development board. Each pixel of the image except the peripheral is filtered by the median filter circuit. We control the SDRAM addresses to extract 9 pixels from the SDRAM. The pixel data list inputs to the median filter circuit. The filter circuit is used to sort the list and selects the middle value of the list. We restore the filtered pixel of the image into another SDRAM and display the image on the screen to show the improved image.