Research On Learning-Based Super-Resolution And SR System Implementation

Super-resolution (SR) is a software technique that enhances the resolution of images or videos by digital image processing. Utilizing SR techniques can improve the resolution of images without updating the existing low-resolution sampling devices; and also can enjoy high-quality videos without increasing the signal transmission bandwidth. Therefore, SR technique has applied in many fields and has the actual research value. This paper researches on SR techniques, including pre-processing technique of SR, learning-based SR methods, and the design and implementation of video SR system.For pre-processing technique of SR, this paper mainly study the problem of mixture Gaussian noise in the image and proposed an adaptive image denoising method for mixture Gaussian noise. Our method combines transform domain with spatial domain, and come up with a block-based and filter-based noise estimation method which can estimate parameters of mixture Gaussian noise adaptively. Then use the estimated parameters to remove noise, gain several denoising images. Through the data fusion technology, achieve the final denoising image. Tests show that our method leads to good results both on noise estimation and noise removal.For super-resolution, this paper focuses on the manifold learning SR algorithm and proposed an improved and fast SR through neighbor embedding method. Based on the framework of LLE, we study the way to recover images rapidly and efficiently. The proposed SR method includes a classification method for the training set based on DCT coefficients, a simple and effective high- and low-resolution feature extraction method and an adaptive and optional reconstruction algorithm.In the experiments, our proposed SR method improves visual quality, meanwhile, the speed greatly. For the design and implementation of video SR system, we design the system with DirectShow framework. The system supports local and camera video sources, offers a variety of SR algorithms, supports previewing the low-resolution video and the recovered high-resolution video at the same time, also supports browsing the videos reconstructed by interpolation and SR method contrastively. And it can save the SR video to the local hard disks.