Scene Registration And Super-Resolution Reconstruction For Infrared Sequence Image

Infrared detection has become more and more indispensable in military and security.However,under the influence of the production process and imaging mechanism of the current infrared imaging technology,IR image has the defects of low globle and local contrast,poor resolution and blurred edge,in addition,the presence of camera jitter during the acquisition of sequence images and the relative movement between the camera and the scene will lead to spatial coordinate transformation of inter-frame images.All these factors seriously limit the further development of infrared imaging.Therefore,this thesis will deeply study the registration and enhancement technology of infrared sequence images.The main contents of our research are as follows:(1)Aiming at the defect of low contrast of infrared image,this thesis mainly studied the enhancement algorithm based on local histogram equalization.First the image was divided into blocks,and SVM was used to classify the image sub-blocks into simple sub-blocks and complex sub-blocks,then unified linear stretching strategy of simple subblocks and histogram equalization of complex subblocks based on the minimum visual error were carried out.Then interpolation was used to solve the block effect caused by the different processing methods of different sub-blocks.The problems of excessive enhancement in the background area and noise amplification were solved effectively.(2)As for image registration,different combinations of common feature detection and feature description algorithms were tested,a feature combination strategy of Harris and FREAK was obtained.In the stage of mismatching elimination,a RANSAC algorithm based on the priori information of sequence images was proposed.Under the premise of ensuring the registration accuracy,the calculation amount of RANSAC algorithm can be greatly reduced,and the real-time performance of the algorithm can be greatly improved.At the same time,combining with the line feature registration algorithm of Hough transform,some image registration problems such as fewer corner points were solved(3)Aiming at the poor resolution of infrared images,the spatial coordinate transformation relationship between the frames of sequence images through the registration algorithm base on the combination of Harris and FREAK was obtained,then combined with the convex set projection method,and the improved PSF,multiple frames of low-resolution images were reconstructed into a high-resolution image.Compared with the traditional POCSbased super resolution reconstruction algorithm,a significant improvement in PSNR and SSIM evaluation indexes were obtained.