Research On Super-resolution Reconstruction Method Of Remote Sensing Image Based On Compressed Sensing

With the development of remote sensing technology,it is more convenient to obtain remote sensing images.However,due to the limitation of the resolution,remote sensing images may not be able to meet the needs of the subsequent processing.In order to improve the image resolution without increasing the hardware costs,super-resolution reconstruction problem has drawn wide attention.In this thesis,we focus on the research about remote sensing image super-resolution reconstruction.Some effective methods based on compressed sensing are proposed by introducing sparse recognition,low-rank decomposition,nonlocal information,and multi-scale.Our work mainly includes the following parts:(1)We study the degradation model of remote sensing images and use the prior information provided by the degradation model to constrain the image super-resolution reconstruction process.First,the image degradation model is constructed according to the degradation factors in the remote sensing image acquisition process.Then,the compressed sensing based image super-resolution reconstruction model is presented.Finally,a joint dictionary is learned to constrain sparse representation coefficients.Experimental results demonstrate that it is feasible to apply compressed sensing theory to remote sensing image super-resolution reconstruction.(2)Classification dictionaries for the typical terrain in remote sensing images are learned and applied to the image super-resolution reconstruction.First,classification features are constructed by computing the radiation-parameters of remote sensing images.Next,the classification of the typical terrain can be determined using sparse recognition.Then,joint dictionaries are learned from different types of training samples.Finally,the test image is reconstructed using the corresponding dictionary.Experimental results show that the proposed method outperforms other image super-resolution methods in terms of the quality of the reconstructed images.(3)The images reconstructed by multiple dictionaries are fused with low-rank decomposition.There is complementary information existing in the images reconstructed by different types of dictionaries.Thus,we perform a low-rank decomposition on the multiple reconstructed images to obtain a fused image which can contain more terrain features and enhance the local details.Experimental results show that compared with the single dictionary method,the proposed method is more effective in terms of the further improved quality of the reconstructed images.(4)A super-resolution reconstruction method combined with compressed sensing and nonlocal information for remote sensing images is proposed.First,the similarity between pixels is calculated according to the structural features of image patches.Then,the weight of similar pixels can be evaluated by merging the local and nonlocal information.Finally,a regularization term combining the local and nonlocal information is introduced into the compressed sensing framework.Experimental results show that the proposed method can better recover the fine textures and effectively suppress the noise.(5)We utilize the multi-scale mode and the compressed sensing framework to remote sensing image super-resolution reconstruction.First,the acquisition process of the low-resolution image is estimated according to the multi-scale sampling method.Then,a discriminative dictionary is learned by multi-scale training sample sets and used for image super-resolution reconstruction.Finally,the nonlocal information of the reconstructed image is used for global optimization.Experimental results show that the multi-scale mode is more effective than the single scale mode and the discriminative dictionary has a good presentation ability.