Image Super-resolution Using Wavelet-domain Dictionary-based

Image super-resolution reconstruction is widely applied in medical imaging, remote sensing images and other fields. Generally, this reconstruction mainly processes noiseless image. For images affected by noise and blur, the independent image de-noising often is required before super-resolution construction. Therefore, how to simultaneously de-noise and improve the resolution has become a hot of the super-resolution reconstruction.At present, de-noising and super-resolution reconstruction methods of image include interpolation methods, learning-based methods. etc. The later methods are the known methods which take advantage of the sparse representation theory to dispose different image conditions such as blur, noise. These methods adaptively adjust parameters to de-noise and reconstruct image without considering that the features information of low-resolution image is easily disrupted by noise. For noising image, the super-resolution reconstruction methods need more operations to deal with it.Therefore, this paper focus on how to effectively use the high-resolution images for synchronization de-noising and super-resolution reconstruction of low-resolution image with sparse theory and wavelet transform. Firstly, the basic knowledge and the theories about the subject are shown and analyzed. Especially, paper summarizes that key technologies of image super-resolution reconstruction, constructors of sparse-based dictionary and basic properties of wavelet coefficient. Subsequently, according to the orthogonality and multi-resolution analysis characteristic of wavelet transform, we propose the wavelet-domain dictionary model, including wavelet analysis dictionary and wavelet learning dictionary, and super-resolution images reconstruction optimization model. Finally, experiments show that the proposed method is better than the existing SR algorithms for de-noising and super-resolution reconstruction of image.