本論文中是以最小平均相關能量理論設計出的濾波器,配合影像位移找出最小平均相關能量的參考函數。在過去的研究中,通常會把訓練影像放在中心位置,但是旁瓣能量不是最小。本實驗利用電腦軟體將彩色圖像先濾成 Y、U、V 三種分量。彩色圖像在範圍 360 度每 5 度旋轉一次,及在範圍 x = -5 至 5 像素, y = -5 至 5 像素進行位移產生多張訓練影像,最後經由 MACE 與互相關計算出每種角度的最小旁瓣能量及其位置所對應的訓練影像並儲存其圖像進行數據分析。根據此法,我們使得輸出平面上的相關高峰值能夠更尖銳及降低旁瓣能量,因而能大大提升了辨識效率及減少判斷錯誤的發生。
This research uses minimum average correlation energy method and shifted training images to recognize the polychromatic images. In our former research, the training images are displayed at the center position. However, the total side lobe energy may not be the minimum. In order to solve the problem, we transform the color image into three YUV color space components. Then, three components are rotated from 0° to 360° in steps of 5° and shifted from -5 to 5 pixels in both of vertical and the horizontal directions to yield training images. Finally, we can obtain the filter with minimum side lobe energy on the output plane by using the minimum average correlation energy method and cross correlation operation. Therefore, we can enhance the efficiency of recognition and decrease the possibility of error.