Research And Implementation On Some Key Techniques Of Ultra-High-Resolution Display System

Combined with the specific application of an Ultra-High-Resolution-Display Wall?UHRDW?system in a command and control center and the requirements of real-time synchronous display of 4K×8 ultra-high resolution?8 times 4K resolution,that is 15360×4320 resolution?,we developed the multi-screen image processor platform architecture and conducted intensive study on hardware platform construction,image enhancement processing,image super-resolution reconstruction and image de-blurring.The key work of this paper includes following aspects: the construction of multi-screen real-time image processing hardware platform,improvement of halo artifacts and contrast loss during the key video frame image enhancement,a non-convex regularized super-resolution algorithm for edge blur and noise in low-resolution degraded image,and the de-blurring of natural images in ultra-high resolution display system.Based on the parallel input of 64 video sources and the real-time display of 4K×8 ultra-high resolution,this paper analyzed the characteristics of current processing platform architecture and built a multi-screen real-time image processing hardware processing platform for ultra-high resolution display system.The system separates the ultra-high resolution desktop image data and real time signal source data on physical space.The external video signal source is normalized to 2.43 Gbps high-speed serial video stream by image acquisition module and the 72×72 space division switching device of system control switching module achieves the system video streaming point-to-point non-blocking transmission to the specified display-output module.Through the computer system PCI-Express bus,ultra-high resolution desktop image data is sent to the specified display output module where the two flows of data are displayed together.The advantage of this platform architecture avoids processing bottlenecks on the PCI-Express bus in existing architectures and enables real-time processing and display.Based on the constructed platform,video segmentation and parallel processing are adopted in the image acquisition module and real-time access processing of 4K super-high-definition signal source?3840×2160@30?is supported.Improvement was made with respect to the defects of halo artifacts and contrast loss during the key video frame image enhancement in the ultra-high-resolution display wall system.Based on the illumination-reflectance model,we used the non-linear diffusion filtering method to estimate illumination and introduced the texture suppression step into the diffusion equation to increase the diffusion rate of the texture region.The proposed method brings about two improvements in the image enhancement application: edge preservation of the original image and loss prevention of texture details in subsequent illumination image compression processing.A non-convex regularized super-resolution algorithm was adopted to solve the problem of edge blur and noise in low-resolution degraded image in ultra-high resolution display system.Using the sparsity properties of l_1/2 regularization and the sparse prior knowledge of heavy-tailed distributions of natural image gradients,a non-convex l_1/2-BTV regularization term is applied to super-resolution reconstruction.Results showed that the proposed algorithm generated high-resolution images of better quality,as defined by both de-noising and edge-preservation metrics.Aiming at the demand of surveillance image high-quality display in command and monitoring application of ultra-high resolution display system,natural image de-blurring is studied.Based on the sparsity of l_p ?0<p<1?regularizer and significantly heavier-tailed hyper-Laplacian model of l_p natural image gradient,we propose an approximation of regularization terms with 2/3?p< 1 using a linear combination of two l_p terms(that is l_2/3 and l₁)with closed form threshold formulae.We used the Least-Squares Curve Fitting Method to determine the weight coefficients.We developed an alternating minimization method to solve the image de-convolution problems involving the constructed approximating function.We derived theoretical analytic solutions and fast closed-form threshold formulae.We performed extensive numerical experiments to demonstrate the versatility and effectiveness of the proposed method through a comparison with the recent non-convex l_p regularization dealing with the special p-value term,with an application to image de-blurring.Our experiments demonstrated that our algorithm becomes more universally adaptable compared to existing specific p-values of the l_p norm?such as,p=2/3?regularization method.In the ultra-high-resolution display system,the subject of multi-screen image processing platform achieves display synchronization and real-time processing,with the realization of 64-source concurrent processing,4K×8 ultra-high resolution real-time display output,and video processing frame rate of 30 fps.Results showed that the system meets the needs of large scale industrial applications.