An Image Registration Method Based On Piecewise Linear Group Parameters

Image registration is one of important research issues in computer vision field, which is the foundation of practical techniques, such as stereoscopic vision, movement analysis, data fusion etc. Image registration is the process to seek one-to-one mapping between images, that is, to relate the points belong to different images at the same position in space. To achieve good experiment result, the key factor is whether we use the proper transformation model. The large body of research into the registration of images, however, has been slanted toward domain-only registration; that is registration between image functions' spatial coordinates. These registration approaches have typically assumed a specific geometric relationship between images' spatial coordinates, for example translational, affine or perspective relationships have been common. Also assumed have been more complex relationships involving the computation of the optical flow field or the modeling of lens distortion. More recently, there is a growing trend in optical imaging involving range-only registration; that is registration between image functions' pixel values. The research seeks to construct high dynamic range maps of an imaged scene from several images captured at different optical setting. The jointly registration of optical images in both domain and range, in order to resolve image registration problem at different optical setting, has been frontier in research field.This thesis, based on comparametric model put forward by Professor Sman, jointly analyses the image registration in both domain and range and do some contributions as follows:(1) A new algorithm, based on joint action of pseudo-perspective model and comparametric function, is proposed to do well in image registration at different optical setting.(2) Further analysis is done to improve the enhancement of piecewise linear comparametric image registration and improvement is also made for range registration details.The key step for us to do analyse of images taken by common cameras at different optical setting for range registration is to calibrate comparametric function correctly. This thesis discusses the application of comparametric function to image registration, and present an improved algorithm for image registration at different optical setting. Compared with other traditional algorithm, this algorithm fully embodies piecewise linear approximation idea, make jointly analysis of images in both domain and range and resolve range difference problem in image registration process. More precise initial parameters of joint domain and rang image registration model for optimal iteration can be obtained with a computationally attractive least squares formalize. That also leads to improve iterative speed. RANSAC and LM algorithm could improve the stability of image registration and the speed of iteration. Final experiments show that this method is doing well to solve the problem of exposure difference between images taken by common camera.In the following chapter, improved analysis of piecewise linear comparametric algorithm is introduced. The introduction solves the problem arising from underexposed or overexposed images registration, serves to reduce error and sensitivity to the number of piecewise linear segments and gain the better result of exposure estimation and piecewise linear approximation. This algorithm's advantage is found as stricter lope constraints in detail and standard selection of knot location.