Super-resolution (SR) has gained widespread attention due to its applications in enhancing image quality, but its computational demands often limit broader usage. This paper proposes a novel framework combining a Preprocessing Layer, an Object Recognition Layer, and a Fusion Layer. The design aims to allocate computational resources more efficiently by processing different parts of the image with tailored methods: the main subject is handled using neural networks for high-quality super-resolution, while other regions are processed with interpolation algorithms to reduce computational load. Extensive experiments conducted on benchmark SISR models and datasets demonstrate that the proposed framework achieves an approximately 30% reduction in FLOPs while preserving competitive visual quality, maintaining high SSIM and PSNR comparable to state-of-the-art SISR performance.