數位相機上的多重曝光影像融合

Abstract

融合多張不同曝光的畫面可呈現具高動態範圍的情境，本論文提出一個影像融合方法，藉由融合兩張連續但不同曝光的視訊畫面，提升畫面的動態範圍，使得在融合畫面中比在典型曝光畫面中更容易找尋移動物件與偵測人臉，提出的方法已實現在具強健硬體平台與軟體平台的消費型數位相機中，由實驗結果顯示影像融合的速度接近每秒四個畫面。 融合多張不同曝光倍率的影像對於在高動態範圍的情境下拍攝靜態影像是十分有用的，然而在實際相機應用面上，由於拍攝者造成相機輕微晃動或是情境中被拍攝者輕微移動皆會使得情境改變，此種情境改變必須進行補償。本論文提出一個完整的影像融合系統，藉由融合三張不同曝光的影像，以達到延展影像動態範圍的目標。不像大部分影像融合演算法均針對被處理過的影像進行融合，且嘗試去還原拍攝系統的轉換函數，我們提出的演算法直接針對還未經過任何影像處理的原始影像資料進行融合。此影像融合系統中亦包含整體性與局部性抖動補償演算法，能有效補償晃動的問題並且得到穩定的融合結果。Fusing multiple frames with different exposure time can accommodate the scenes with high dynamic range. In this thesis, we propose an approach that is to fuse two consecutive video frames with different exposure time. Finding moving objects and human faces in such a higher dynamic range fused image is much easier than the typical exposed frame. The proposed approach has been implemented on a commercial digital camera with robust hardware and software platform and the experimental result shows that the fusion speed is around 4 frames/seconds. Fusing several differently exposed images is particular useful for taking pictures in high dynamic range scenes. However, the scene changes resulted from moving objects and vibrations caused by photographers must be compensated adaptively in practical camera applications. In this thesis, we propose a complete image fusion system aiming at extending dynamic range of a picture by fusing three differently exposed images. Unlike most of fusion algorithms operate on processed images and try to recovery the transfer functions of imaging systems, the proposed image fusion algorithm directly works on raw image data before performing any color image processing. The proposed global and local stabilization algorithms efficiently remedy the vibration problems and achieve a quite stable image fusion result.