Abstract
在準確度要求不高的情況下，以像素大小的解析度為基礎所定義出邊緣的位置，在影像處理上以可以滿足其需求。然而諸如照像量測、工業檢驗等，對於準確度的要求相當高，一般的準確度並不夠，所以次像素技術被用來解決這問題。整個論文除了以準確度問題為重心外，還要發展發展新的技術以解決機器視覺在製造現場所遭遇到的難題。本研究提出兩個次像素技術：LOG 旋積之線性內插法 (Linear Estimation for LOG Operationmethod)與加權平均法 (Weighted Average method)。LOG 旋積之線性內插法將影像以LOG 遮罩作旋積處理後，利用線性內插去定位影像邊緣；而加權平均法則是將相鄰兩像素之灰階強度差定為加權之權數，再依此一加權平均值所對應之位置定為邊點的估計位置。為了有效地驗證本研究所提出之次像素技術，我們合成多張大小為100x100 像素的影像來作測試。實驗結果顯示LOG 旋積之線性內插法所定位之邊緣位置比加權平均法來得準確。然而LOG 旋積之線性內插法對於雜訊之敏感度比加權平均法高了許多，且加權平均法在有雜訊干擾下，仍有不錯的結果；再則，加權平均法的處理時間約只有LOG 旋積之線性內插法處理時間的六十分之一。最後，我們同時發展出一個套統檢驗整合軟體來說明機器視覺在套筒檢驗應用上的及時檢驗之能力。The employment of a machine vision system for part inspectionhas become an important component for achieving the goalof real-time control. However, one of the majorconstraints of this technique is its limited accuracyowing to the pixel resolution. For those applicationswhich require better accuracy, such as photogrammetry andprecision measurement, the accuracy can be improved byreplacing with better hardware devices or by applying thesubpixel techniques. In this research, we intend to developnew machine vision techniques to solve the problemsencountered in practical manufacturing environment, withemphasis on accuracy problem. Linear estimation for LOGoperation (LEFLO) method and weighted average (WA) method areproposed in this research. The LEFLO method applies theLaplacian of Gaussian convolution and localizes the edgewith linear interpolation. WA method uses the deviation ofgray values of successive pixels as the weight numbers toestimate the gray level of the edge translation. In thisresearch, the synthetic images are employed to evaluate theeffectiveness of the algorithms. The experimental resultsshow that LEFLO is more accurate than WA when images arenot disturbed by the environmental noise. Nevertheless,LEFLO is more sensitive to noise. The weighted averagemethod requires less processing time than LEFLO method.It is almost one-sixty of the processing time of the LEFLOmethod. Finally, an integrated system for socket inspectionis developed to illustrate its capability real-timeinspection.