針對平移與旋轉晃動偵測與消除之數位式影像穩定設計

Abstract

畫面晃動的現象普遍存在於視訊當中，而成因主要來自於攝影機的搖晃。這些搖晃通常是些高頻且不平順的運動。數位式影像穩定化設計的目的便在於消除這些畫面晃動的現象。數位式影像穩定設計主要可以分成兩部分，包括運動的估測和視訊的補償。運動估測的目的在於利用影像處理的技術來估測攝影機的運動行為。視訊補償的目的在於將所估測的運動行為中不必要的運動移除，提供穩定的視訊。 本篇論文主要針對平移和旋轉的畫面晃動提出演算法。運動估測的準確性常影響著影像穩定的效果。當攝影機同時受到平移和旋轉的搖晃時，為提昇運動估測的準確性，本論文提出一個尋找攝影機運動的旋轉中心的概念，且根據這個概念推導出能獲得平移、旋轉和旋轉中心等相關資訊的仿射運動模型。

The task of image stabilization systems is to eliminate the image fluctuations caused by unwanted camera movement, while preserving only the intentional camera movement. A digital image stabilization system can be divided into two major parts: one is the motion estimation unit which aims at estimating the global motions of inter-frames and another one is the motion correction unit which aims at eliminating image fluctuations. In this thesis, an algorithm of digital image stabilization which aims at handling both translational and rotational fluctuations is proposed. In order to improve the accuracy of global motion estimation, a concept based on searching the rotational center of a rigid camera motion is proposed and a modified 3-parameter affine motion model is derived. Then, information about the rotation angle, the rotational center and the translation vector can be obtained.