以內向式雙鏡頭取像之影像縫合技術

Abstract

本篇論文主要探討在雙鏡頭時，內向式和外擴式的兩種擺法所造成的差異，近年來影像縫合越來越熱門，因為大家都希望能夠得到更大視角的影像，所以我們透過相機焦點距離、縫合後的像素差距、相機最大視角還有重疊區域，分別去討論和比較這兩種擺法所得到的結果，發現內向式的影像縫合可以大幅降低”鬼影”現象，而且內向式影像縫合比外擴式影像縫合有著較大的重疊區域，如此可以增加特徵點的匹配數，進而加強縫合的準確性。除此之外，之前的演算法在近距離縫合時會出現問題，主要是當近距離拍攝影像時，影像容易出現明暗不一，會造成原本的演算法在匹配特徵點時出現錯誤，所以我們使用不同的方法來解決影像明暗不一時無法縫合的問題，我們主要用相關性、像素值排列還有邊緣判斷三種不一樣的方法來試著改善此問題，也比較了這三種方法的效果和速度。最後決定使用邊緣判斷，因為他的速度比其他兩種方法都快了2-3倍，而且也能夠縫合大部分的圖。

This thesis aims to explore the efficiency of two approaches for dual-camera image acquisition. There are two setups for dual-camera system, considered in this thesis. One is the inward image sensing (IIS) and the other is the outward image sensing (OIS). Recently, image stitching is more and more popular. Everyone wants to get image which has wider view angle, so we discuss and compare the result of two arrangements by their distance of focal point, the disparity after stitching, the widest view angle, and overlapped area. We found that IIS can decrease the phenomenon “ghost-effect.” Otherwise, IIS has wider overlapped area, it can increase the matching point of feature points, and increasing the accuracy of image stitching. Besides, there is a problem when we stitch image at short range in previous algorithm. Because there is different brightness in two images when we take pictures at short range. It will make mistake when we are matching feature points in previous algorithm. So, we use some different method to solve the problem. We mainly use “correlation matching” ,”pixel ranking” and “edge enhancement” to improve the problem. And also, we compare the effect and speed between the three methods. Finally, we choose “edge enhancement”, because it has two or three times speed than other method and it can stitch most photos.