Research On Multi-band Sea Surface Image Registration And Fusion Method

With the continuous development of short-wave and long-wave infrared technologies and the expansion of their application fields,the limitations of single-band imaging are gradually emerging,making it difficult to meet industrial needs such as security surveillance and target detection.Long-wave infrared images are imaged by accepting infrared radiation from the target and the background,enabling all-weather detection and better differentiation of target and background information.Short-wave infrared imaging enables image acquisition through fog,and have great advantages for application in maritime scenes.Visible light images are rich in detail with higher resolution,and are more in line with human visual perception.The fusion of multi-band maritime images supports the synthesis of the information of each band to enhance the mastery of the target and the understanding of the background.In addition,allweather and multi-course monitoring of sea surface conditions can be achieved through the fusion of multi-band images.In general,this thesis firstly introduces the theory of image registration and image fusion,and analyzes the characteristics of multi-band images of maritime scenes,and proposes a registration method for long-wave infrared and visible images based on the main direction of contour features.The method first extracts the contour of the image and determines the main direction of the contour by extracting linear features based on the contour image,and then completes the registration of the two images based on the feature description of SIFT(Scale Invariant Feature Transform)method.In the process of extracting contours,the contours of the sea surface region are rejected to avoid feature extraction in this region.The main direction determination method and the rejection of the contour of the sea surface area effectively reduce the computational effort,reduce the probability of mis-matching,and improve the speed and accuracy of registration.For the short-wave infrared and visible images,the descriptors are improved in this thesis based on the SuperPoint method.The thresholding normalization operation of the descriptors improves the image registration effect.In this thesis,the visible image is used as the reference image,and the registration between the long-wave infrared and visible images and between the short-wave infrared and visible images is accomplished indirectly for the purpose of registration between the long-wave infrared and short-wave infrared images by two-by-two registration between the long-wave infrared and visible images and the short-wave infrared and visible images.For the aligned three-band images,a fusion method combining visual saliency and multiscale decomposition is proposed in this thesis.The method uses a guide filter to decompose the three-band image into multiple scales to obtain the corresponding detail layer image and the base layer image for the first time.For the detail layer image,the maximum principle is used for fusion,and for the base layer image,a fusion strategy based on the grayscale weighting of the saliency map is used,and the texture structure and grayscale information of the image are effectively retained in the final fusion result.