| 摘要: | 本研究目的在於以指數校正法改善陰影對於聚焦成形(Shape From Focus)技術所帶來的影響。聚焦成形技術主要應用於量測物體的表面形貌及三維重建，可全場性的量測物體。在機器視覺取像的過程當中，往往會因為外在環境的因素，如照明、震動等，進而導致拍攝出來的影像品質不佳。而本研究將針對陰影所帶來的影響做改善。
聚焦成形(SFF)的原理是將待測物放置於參考平面上，經由移動已知的距離，使CCD或待測物位移，位移過程中不間斷的重複擷取影像，以不同序列影像的清晰程度去判斷對焦時當下的位置，進而推算出深度圖。主要就是根據每張影像序列中鄰域內光強度的差異，判斷出其表面形貌的清晰程度，清晰的程度是以聚焦度值演算法計算而來。同鄰域中不同的影像張數，在模糊時的光強度差異小，而當影像清晰時，光強度差異大。本研究經實驗發現表面存在陰影時，會降低影像的光強度，造成表面平滑或缺少紋理，導致系統無法判別出每張影像的差別。因此，本研究提出一種指數校正法，能夠有效的改善陰影所帶來的影響。在實驗中，我們使用多種的演算法作為計算聚焦度值的方法，並用所提出的指數校正法對影像進行校正，分析校正前後的差異。透過實驗驗證指數校正法的可行性及改善能力，且利用白光干涉儀驗證本研究的聚焦成形系統之量測性能。其系統的空間解析度為1 m/pixel，軸向解析度介於50 ~ 100 m 間。以及指數校正法能夠降低至少10 %以上的誤差。;The purpose of this research is to improve the effect of shadows on Shape From Focus(SFF) with the exponential correction method. The SFF is mainly used to measure the
surface topography and three-dimensional reconstruction of objects, which can measure objects in full field. In the process of machine vision, external environmental factors, such as lighting, vibration, etc., often lead to poor image quality. This paper will improve the impact
of shadows.
The principle of SFF is to place the object to be measured on a reference plane, move the CCD or the object to be measured by a known distance, and continuously capture images during the displacement process, judge the focus based on the clarity, and then calculate the depth map. It is mainly based on the difference in light intensity in the neighborhood of each image to determine the value of focus of its surface topography. The value of focus is calculated by the focus algorithm. In this study, we found through experiments that shadows on the surface will reduce the light intensity of the image, resulting in a smooth surface or lack of texture. As a result, the system cannot distinguish the difference between each image. Therefore, before using the focus algorithm to calculate the image, the image is corrected. We use a variety of algorithms to calculate the focus value, and experiment with or without shadow on the surface. We use the exponential correction proposed by this research to correct the image and analyze the difference before and after correction. The measurement performance of this system is verified by a white light interferometer. The spatial resolution of the system is 1 um/pixel, and the axial resolution is between 50 ~ 100 um, and the experiment correction can reduce the error by at least 10 %. |
