數位相機影像分析中的幾何校正、景物疊合及物體分離

Abstract

當數位相機在取像時，可能無法把同一個景物放入單一的影像裡，如果經由調整焦距，強迫把此景物完全拍攝入單一影像裡，則此影像的解析度勢必被降低。為了顧及影像的解析度，本論文提出了一個藉由不同的取像角度拍攝多張影像，再以影像疊合技術重建該景物的技術。在重建景物前，這些影像先經過幾何校正，以消除因相機鏡頭取像時所引起的幾何扭曲，再採用通用式赫夫轉換(generalized Hough transform)的比對方法，以獲得待疊合影像的幾何轉換關係，並且提出一個色彩校正方法，用以校正欲疊合影像間因取像時間不同所造成色彩及亮度的差異。此外，我們也提出一個在物體邊界外，藉由輔助的手繪封閉輪廓線，以抽取物體的方法。實驗結果良好，證明了所提方法的可行性。

When a scene is taken from a digital camera, it may not be entirely included in the taken single image. Also, if the scene is forced to be included in the image by adjusting the focal length, the resolution of the scene in the image will be decreased. This gives us the motivation of studying the use of multiple images to construct a scene. A new approach for this purpose is proposed. The images used for scene construction are taken from different camera views such that the scene is entirely covered by these images. Before the construction process, geometric correction is conducted on these images first to eliminate geometric distortion in these images, which is caused by imprecise camera lens geometry. Then, an iterative generalized Hough transform is proposed to obtain corresponding matching points within these images. The matching points are then used to stitch these images for scene construction. To avoid serious color differences in these images caused by image taking at different times with different brightness, a color correction technique is also proposed, by which the constructed scene can be modified to look smoother. Additionally, a semi-automatic object extraction method is proposed. A close contour of an object is drawn manually by the use of the computer mouse. Then, by using the Sobel operator to track the boundary points of the object, the object shape boundary is extracted within the close contour. The method is applicable even when the background of the object is complicated. Good experimental results are also shown to prove the feasibility of the proposed approaches.