| Summary: | 碩士 === 國立中興大學 === 機械工程學系 === 88 === Subpixel edge detection method in machine vision mainly includes edge model and edge estimation method. Most of the conventional one-dimensional edge estimation methods use the ideal step or the two-sided exponential function as edge model and the center of shape method or the least-square method as edge estimation tool. Subpixel edges are estimated by one of the four cross combination approaches. When the images obtained are blurred or out of focus, the performances of these methods degrade tremendously.
In this thesis, the bipolar continuous function and the correlation method are adopted as new edge model and subpixel edge estimation approach, respectively. Performances comparisons among the four possible subpixel edge estimation methods are carried out by real experiments to improve the accuracy and robustness.
To further reduce the computing time of the one-dimensional methods, two new two-dimensional subpixel edge estimation methods are proposed in this study. In these two new approaches, the two-dimensional two-sided exponential function is chosen as the edge model. The two-dimensional least-square method and the correlation method are the estimation tools, respectively. Experimental results show that all the 1-D and 2-D estimation methods proposed outmatch the conventional methods both in system stability and computing time.
|
|---|
