摘要
本論文所探討的問題為，在相機為靜止的情形下，如何去除物體因移動而產生的模糊。在快門打開的期間，運動物體與靜止不動的相機間之相對運動，會造成物體模糊的結果。除了物體本身變得模糊外，物體的邊緣也會混入部分的背景。多數去除移動模糊的方法是針對整張影像進行去模糊。如此，雖然運動物體本身因而會變得清楚，但原本清楚的背景卻反而會變得模糊。又因在運動物體的邊緣處混入了背景的成份，故去模糊的效果在物體的邊緣處將不會很好。要能夠將混入物體的背景去除乾淨，我們在此需要假定物體在靜態時的輪廓為已知。 在本方法中，首先我們使用中間值濾波法，從影像序列中擷取出不含物體的純背景。再藉著比較含運動物體之影像與純背景影像，在顏色上與梯度上的差異，從含運動物體的影像中，找出運動物體的大致輪廓。而物體在影像中的位移量，可經由交疊運動物體之輪廓與靜態物體之輪廓估計出來。另外，造成模糊的PSF(point spread function)可從單獨的實驗中推出。再來我們利用位移量及找出之PSF，對分離出的運動物體去除混入的背景成份，如此即可得到不含背景成份的運動物體。接著使用Wiener Filtering來還原物體。最後，將還原後的物體影像貼到背景影像上，就可得到物體與背景皆清楚的還原影像。實驗結果顯示此一方法可以有效的達到去除物體本身的移動模糊。
並列摘要
The main issue discussed in this thesis is on how to restore a blurred object in an image due to motion. Here the camera is assumed to be stationary. During the period that the shutter of the camera is open, the relative motion between the object and the camera causes the object to look blurry. Besides the object being blurry, certain amount of the background is also blended into the object. Most approaches apply the de-blurring process to the entire image. In this way, though the blurred object becomes clearer, however, the originally clear background will become blurry. Also because part of the background is mixed with the object, the de-blurring result usually are not very good at the leading and trailing edges of the object. To be able to remove the mixed background entirely from the object, we need to assume that the shape of the object is known. In our approach, we first obtain the pure background image from an image sequence using median filtering. Then we compare an image with a blurred object with the pure background to extract the silhouette of the moving object. The comparison is made on both their differences in colors and their differences in the gradient of the colors. Then by overlaying the silhouette of the static object with that of the moving object, we are able to derive the displacement of the moving object in the image. Also, we can derive the PSF of the blurring effect in a separate process. With the knowledge of the displacement and the PSF, now we are able to remove the part of the background that is mixed into the object. We use Wiener filtering to de-blur the object and paste the result onto the static background to obtain a restored image with both the object and the background are now better defined. We experimented with images with objects that have different degrees of blurriness, and have obtained satisfactory results.