Low-light Images Registration Algorithms

As a fundamental problem of image processing, image registration refers to match two or more images at different times, from different se nsors or from different viewpoints, and its aim is usually to provide a more precise understanding of the target for the follow-up image processing. Image registration techniques has a very wide range of applications in many areas, such as remote sensing image processing, medical image processing, computer vision and pattern recognition.. The low-light image registration is also a typical use of image registration, and has a broad practical field. However, the existing registration algorithms often cannot achieve a satisfactory result when dealing with low-light images, resulting in many image processing applications in low-light scenes have not been very good development. Therefore, the study of low- light image registration techniques will not only help to enrich the image registration theory in low- light environment, but also help to open up the range of image registration applications, and has important theoretical significance and high application value.Firstly, after analyzing and summarizing the trad itional area-based methods and feature-based methods, for their shortcomings and the characteristics of low-light images, and melting area-based methods among the framework of feature-based methods, we propose a new low- light image registration method based on block matching. The method uses block matching result of filtered images to obtain the corresponding relationship between the registering images, and then establish the image transformation model, and thereby complete the image registration work. In order to verify its effectiveness and superiority, the paper conducts a comparative experiment on registration accuracy and registration time. Simulation experimental results show that the method expands the scope of registration algorithms to low-light images. At the same time, it not only solves the sensitive issue of traditional area-based methods, but also plays much better than feature-based methods on registration speed.Secondly, in order to further improve the accuracy of registration, based on the proposed method, the paper presents a new block matching method combined withfeature points. According to the research and analysis KAZE feature extraction algorithm inspired, we proposed to use KAZE algorithm to obtain the critical and non-critical areas of an image in order to remove inaccurate pairs from block matching collection, and gain more exact correspondence between the sensed image and the reference image, and finally converted to a more accurate registration result. Simulation experimental results show that the proposed method can further improve the accuracy of the low- light image registration, and has a faster registration speed comparing with registration algorithm based on KAZE feature. In addition, the method solves the difficulty for setting detection threshold of feature-based methods, and is a new efficient and stable image registration method.