Stripe Noise Removal In Remote Sensing Images By Variational Methods

Linear array sensors are the most commonly used imaging devices in optical remote sensing imaging.However,due to the constraints of the system performance or the impacts from the external environments,the imaging detectors may work inconsistently in the scanning imaging process,and result in the appearance of the alternate bright and dark scans in the captured images,known as stripe noise.This kind of noise is different from other kinds of noise in remote sensing images.It possesses the significant structural characteristic,which can make a destructive effect on the original image features,and further limits the image interpretation and the subsequent applications.Therefore,it is important and valuable to undertake the research on destriping.Considering the stripe cases are changing and complicated,whereas the most existing methods are lack of universality and robustness,this dissertation aims at constructing the high-performance destriping models based on the characteristics of stripe noise by using the rigorous and flexible variational framework.The main contents are shown below with the main line of processing sparse stripes,multiple types of stripes,and oblique stripes.A variational model that takes account of the global and local stripe features is proposed to achieve the accurate separation and removal of sparse stripe noise in remote sensing images.On the basis of sparse theory,this approach completes the precise depiction of stripe noise via joint sparse constraints.At the global scale,the sparse distributional property of the stripe noise is used to impose sparse constraints in the spatial domain to "lock" the specific location of the stripe noise.While at the local scale,the asymmetric directional characteristics of the stripe noise is employed to construct the sparse constraints in the horizontal and vertical gradient domains to describe the local details of the stripe noise.Experiments show that the method can achieve accurate removal of sparse strips without needing to obtain the stripe position in advance,and can also conquer the over-smoothing problem of the existing methods.A one-dimensional-filtering-guided universal destriping framework is proposed to overcome the complex degradation of the multiple types of stripe noise.Based on the multi-angle analysis of stripe noise characteristics,this model first predicts the ideal one-dimensional mean curve from the two-dimensional noisy images,and then uses the prediction as guidance information to constrain and control the destriping.In order to ensure the accuracy of the guidance information,the estimation of the ideal mean curve(through one-dimensional filtering)well adapts to the strip noise with different distribution densities.Experiments under different types of horizontal and vertical stripe cases confirm the good robustness of the proposed model.Even under the condition of high-density stripe noise,reliable restoration results can still be calculated,which solves the problem of insufficient universality of the existing methods.An oblique stripe processing model based on oriented variation is proposed to achieve high-fidelity removal of oblique stripe noise with arbitrary directions.This method starts from the strip orientation,and calculates the direction information of the stripe noise by combining the Fourier transform and the guide filter.Then,from an image decomposition perspective,a high-precision stripe separation technology well considering the stripe and image features is designed.To model the oblique stripes,the orientation information is imbedded into the oriented variation to capture the along-stripe smoothness,while an l1-norm term is used to describe the property of the global distribution of the stripes.As for the latent clean data modeling,the total variation(TV)technique is employed in the proposed method.Experimental results show that the method can accurately calculate the stripe directions and achieve the high-precision separation of oblique stripe noise with different directions.It breaks through the bottleneck in the prior art on oblique stripe removal techniques to a large extent.