Weighted quantization and hamming search for fast image super-resolution

Image super-resolution (SR) is a problem estimating high resolution image according to low resolution image. A number of practical approaches have been proposed ranging from interpolation-based to neural networks. In this paper, we focus on the patch-based neighbor embedding approach and propose a fast weighted quantization and Hamming search (WQHS) algorithm. At the offline stage, the WQHS method jointly pursue the linear projections for binary coding and the corresponding weight coefficients, which together can largely reduce the binary quantization loss. Based on the learnt hash functions, the database patches of low resolution can be indexed using the multiple hash tables. At the online stage, we devise a fast nearest neighbor search strategy for each query patch of low resolution that can work well with the code weights over the indexing tables. We evaluate our method on standard image datasets and demonstrate competitive or even better performance, compared to the state-of-the-art methods.