Research On Image Registration Based On Motion Information Of Sensor

Image registration is the process of finding the matching relationship between two images of a same scene.Image registration plays an important role in image quality improvement,remote sensing image analyzing,and medical image processing as a fundamental image processing technique.With the increasing popularity of photography,the demand for faster running speed and higher accuracy of image registration algorithm grows up.Unlike the popular feature-based approaches,which use information only from the images,we utilize the motion sensor which is equipped in most imaging device to registrate images.The proposed method can accelerate the matching process which is meaningful for the high dynamic range images and multiframe image denoising.This paper firstly introduces the background of our research and the applications of image registration,and summarize the current status of this research.The basic theory and related approaches are also introduced,including the basic mathematical model of image registration,the approaches based on gray scale,transformation domain,feature.We summarize the implementation steps of these approaches.In this paper,image registration method based on motion information of sensor is proposed.Based on the camera projection model and the transformation of camera's motion,we derive the model of image coodinates changes with camera's rotation angles.To deal with the measurement error of rotation angles,we propose an accurate matching method for a wider application range.We set a searching window for feature point matching,the window size is related to the measurement error of rotation angles.Thus we can avoid global search in feature point matching process,and accelerate the matching speed.Then we use the set of matching point to fitting the image transformation matrix,based on RANSAC(Random sample consensus)method.We design a hardware verification scheme for the image registration method above,and build the hardware platform.We use the MMES(Microelectromechanical systems)gyrocope to acquire the motion information of camera,the gyroscope is mounted with the camera rigidly.Then the two images are registrated according to the rotation angles between their exposure moment.The experiments show that,our methods is effective and compared to brute force matching,our method reduce 31%running time.