Research On Fast Registration Of Spatial Remote Sensing Infrared And Visible Light Images

Image registration refers to the process of transforming multiple images acquired from different sensors or at different time into the same coordinate system.Remote sensing image registration is an important step in the utilization of remote sensing image information and is the prerequisite for subsequent remote sensing image processing.As technology and crafts of manufacturing sensor become more and more mature,the weight and volume of sensors are greatly reduced.It is feasible to integrate multiple imaging sensors into the same system,and multiple data resources can be obtained at the same time.Combining these data can effectively improve the utilization of remote sensing images.That makes remote sensing resources play a greater role.At present,spatial remote sensing images play an important role in many fields such as environment monitoring,military reconnaissance,and disaster relief.In remote sensing imaging,the imaging mechanism of visible light and infrared sensors is different.The visible light sensor mainly detects the reflected light of the object.The visible light image generally has higher resolution,and the details of the object in the image are clear.However,the visible light sensor is susceptible to the influence of natural conditions,such as weather and illumination.The infrared sensor detects the heat radiation of object that reflects the temperature and material information of the object,and can work day and night.Therefore,the combination of visible light sensor and infrared sensor can not only enhance the complementarity between scene information but also meet the needs of all-day monitoring.Due to the different spectral characteristics of the infrared and visible light bands,the imaging results of the same scene show a large difference.And because of the particularity of remote sensing images,there may be interference factors such as partial occlusion and illumination changes,and the sharp increase of the amount of remote sensing image data makes the registration of infrared and visible light remote sensing images very challenging.This paper first introduces the significance of visible light and infrared image registration technology,and summarizes the research status and basic methods of image registration algorithm.In line with the imaging characteristics of visible light and infrared remote sensing images,the differences and similarities between the two are analyzed,and the registration algorithm between them is deeply studied.The main research work is as follows:1.A fast image registration method based on binary feature description is proposed.The method can improve the registration speed between remote sensing images with small gray scale differences,and combines with hardware implementation at the beginning of design,fully considering the specific hardware implementation method of description and matching process.On the basis of parallel computing,description and matching directly in binary form makes the processing of the algorithm fast and efficient.This method uses FAST algorithm as the feature point extraction algorithm,and proposes a novel binary description and matching algorithm.The polar coordinate system is constructed with the feature points as the center,and the circumference of the feature point neighborhood is divided into several sector intervals.Each interval is divided into several smaller sampling intervals according to the distance from the center of the circle,and the pixel intensity of the small sampling interval in the same sector is recursively compared and binary coded as a descriptor of the feature point.At the same time,the sum of the intensity difference of adjacent sampling intervals in the same sector is counted as the strength of feature point,the direction of the sector with the largest sum is recorded as the main direction of the feature point.The cyclic shift is used in normalization of the main direction to solve the image rotation problem.Compared to traditional descriptors,the binary descriptor calculation is much simpler.Filtering the feature points to be matched according to the strength of the feature points also improves the computational efficiency of the algorithm.The normalization of the main direction of the feature points is performed by means of cyclic shift,which also reduces the complexity of the algorithm.The experimental results show that the descriptors constructed in this way is convenient and fast for the calculation process and is highly robust to the image rotation.2.A method of registration for infrared and visible images based on edge complexity is proposed.In order to suppress the difference in feature point description between the infrared and visible images due to resolution and gray scale difference,the image is converted to the frequency domain,and the obtained edge consistency response function value is converted into an image for registration.Because the edge obtained by the extraction algorithm is basically consistent with the visual effect of the human eyes,the difference between the infrared and visible images is suppressed to some extent,so the obtained edge images only have the difference in the resolution and the local intensity.However,due to the difference between resolution and local intensity,the localized regions with rich texture in the edge image have the problem that the feature points are dense that cannot be correctly matched.In order to further improve the accuracy and efficiency of the registration algorithm,a method for calculating image complexity is proposed to shield these complex regions.With no feature point pair that has a large difference because of the resolution and gray scale difference extracted in these regions,it helps to reduce the number of this kind of point pairs.The experimental results show that the image registration algorithm based on edge complexity proposed in this paper effectively suppresses the difference between infrared and visible images,reduces some invalid feature points in the image,and greatly improves the speed in feature extraction and matching process.It also improves the robustness and accuracy of the algorithm.3.A registration algorithm for visible light and infrared images based on global structural features is proposed.The algorithm can be applied to the registration of visible light and long-wave infrared remote sensing images with greatly different gray scale and resolution.The algorithm constructs feature point descriptor based on the relative distribution structure between feature points,and does not depend on the neighborhood gray of the feature points,so that the influence of gray scale difference between images can be suppressed.The optimized SURF algorithm is used to extract the maximum intensity points of the two images to ensure the repetition of the extracted feature points in the two images,which can improve the accuracy of the algorithm.The polar coordinate system is constructed with the maximum interest intensity point as the origin,and the feature points are binary coded and matched according to the positions of the remaining feature points in the polar coordinate system.Through the construction of the feature points by the relationship between global structure descriptors,this method not only solved the gray scale difference between visible light and long wave infrared registration difficulties,but also continued to take advantage of low computing complexity of the binary descriptor.Moreover,with the method of loop matching,the rotating performance of the algorithm has high robustness.The experimental results show that the descriptor constructed in this way not only improves the registration speed,but also completes the registration of long-wave infrared and visible remote sensing images with low repetition rate of the homologous points due to large differences in grayscale and resolution.4.An embedded system capable of realizing infrared and visible image registration algorithms is designed.Aiming at the requirement of rapid registration of infrared and visible images in engineering,the image registration algorithm is transplanted in the embedded system,and the process of image acquisition-registration-splicing is outputted by the "FPGA+double DSP" framework.The FPGA receives the images,then distributes the visible image and the infrared image to two DSP chips for feature point extraction and description,and then receives the feature point descriptors constructed by DSP for matching.Benefited by the powerful parallel processing capabilities of embedded systems,fast image registration is achieved.