A fast CIS still image stabilization method without parallax and moving object problems

Abstract

In this paper, we propose a fast still image stabilization method for CMOS image sensors (CIS) which can solve the parallax and moving object problems in image registration. CIS still image stabilization is a software technique that prevents motion blurs caused by minute shaking of a CIS camera, and it is realized by seamlessly registering several rapidly exposed CIS images. Our method consists of two stages of image registration: a global registration and then, a local registration. Since a CIS image can be distorted by the motion of a CIS camera having a rolling shutter mechanism, we present a way of removing CIS distortions in the global registration. Also, the local misalignments are reduced using a mesh-based local registration. Since the parallax or moving object regions cannot be correctly aligned in the image registration process, we propose a two-stage post-processing method that removes the misaligned regions by measuring the registration quality based on the labeling method. The computational load in the image stabilization is reduced by selectively performing our local registration based on a real-time global motion estimation method. The experiment results show the necessity of CIS distortion analysis in the CIS image registration and the effectiveness of our stabilization method even for the moving object and parallax problems(1).