Research On Sparse Representation Based Video Super-resolution Reconstruction Algorithm

The purpose of image super-resolution reconstruction is to reconstruct a high-resolution image from a single low-resolution image or multiple low-resolution images using signal processing technologies. Image super-resolution reconstruction algorithms can be classified into single image super-resolution reconstruction and multiple images super-resolution reconstruction. This paper focus on image super-resolution reconstruction algorithm which uses several low resolution frames to construct a high definition one of the current frame. Sparse representation based image super-resolution reconstruction technique is a hot topic in the field of image processing. The key technologies include sparse representation model of video images, dictionary construction algorithms and motion estimation. Based on the analysis of existing dictionary construction algorithms’performance and the effect of the training sample set on reconstruction quality, a sparse representation based video images super-resolution reconstruction algorithm is proposed. The main works are as follows:1)The sparse representation based video super-resolution reconstruction model is proposed. On the basis of sparse representation model of a single image, the sparse presentation based video images super-resolution is proposed, and the solving algorithm of the model is designed.Experiments show that this algorithm can improve the PSNR and SSIM of the reconstructed images significantly.2) The influences of the training sample sets and dictionary construction algorithms on the performance of super-resolution reconstruction are numerically analyzed. Two typical dictionary construction algorithms are implemented. Three kinds of experiments on synthetic and real images show that, the feature based method can achieve better performance than gray value based method on the aspect of peak signal to noise ratio (PSNR) and structural similarity (SSIM), and it also show that the higher similarity between the training sets and the reconstructed image is, the higher PSNR and SSIM can be achieved.3) A multi-scale auto-convolution feature matching based implicit motion estimation method is proposed. This method combines the MSA feature matching and Lucas Kanade method to achieve registration between image blocks, which takes advantage of the invariances of image block such as rotation invariance, translation invariance and scaling invariance. Experiments show that the accuracy of implicit motion estimation is improved, and the PSNR and SSIM of the reconstruction images are also improved.Numerical and visual experimental results show that the proposed method can achieve better performance on visual quality, and higher PSNR and SSIM value compared with existing methods.