Enhanced Single Image Super-Resolution using Modified Very DeepSuper Resolution Network - Gaussian Mixture Models

Abstract Single Image Super-Resolution (SISR) aims to enhance the quality of low-resolution images by reconstructing their high-resolution counterparts. This paper proposes a hybrid method, termed mVDSR-GMM, which leverages the strengths of both Gaussian Mixture Models (GMM) and a modified Very Deep Super-Resolution (mVDSR) network. The mVDSR-GMM method utilizes GMM to model the statistical properties of high-resolution images, enabling accurate reconstruction of high-frequency details. The enhanced mVDSR network, with increased depth and Leaky ReLU activation functions, further boosts the model's learning capability and robustness. Through extensive evaluations using satellite imagery datasets and metrics such as PSNR, SSIM, BRISQUE, and MSE, mVDSR-GMM consistently outperformed existing techniques, demonstrating its effectiveness in improving spatial resolution. Notably, the hybrid method showed superior classification performance in land use/cover analysis compared to the traditional Bicubic method. The proposed mVDSR-GMM emerges as a promising tool for advancing satellite imagery resolution, with potential applications in remote sensing and geospatial analysis.