A Novel Image Super-Resolution Technology Based on the Wavelet Coefficients Prediction Scheme

The super-resolution technologies can be roughly categorized into the sparse coding based methods and learning-based methods. In the sparse coding based methods, the image patches or wavelets coefficients are often utilized to establish the image patch database. However, the time-consuming searching process makes the real-time applications difficult. In the learning-based methods, the DFT or DWT coefficients are used to train the machine learning system. In this paper, we propose a novel learning-based super-resolution method with wavelet coefficients prediction scheme to rebuild the high-resolution images with high PSNR and SSIM scores. The license plate and ecological duck images are used to analyze the performance of the proposed method. The experimental results show that the reconstructed high-resolution license plate images can have PSNR 48 dB and SSIM 0.99 and the reconstructed high-resolution ecological duck images (with high texture distribution) can have PSNR 33 dB and SSIM 0.98. The proposed method outperforms the conventional methods in terms of PSNR and SSIM. Furthermore, the efficiency of the proposed method is fast enough for the real-time applications.