Low-quality images often significantly impair the effectiveness of most computer vision tasks. Improving image quality is challenging due to the diverse range of image types and degradation properties. In this paper, we examine the limitations of three image enhancement approaches based on traditional algorithms and deep learning and enhance them to achieve a better trade-off between the reconstruction quality and time cost. Specifically, we investigate advancements in the receptive field, lightweight design based on distillation, and residual learning for enhancing image brightness, correcting chromatic aberration or color cast, and deblurring images, respectively. Besides, we further evaluate the effectiveness of our image enhancement results by applying them to the downstream task of image segmentation. Extensive experiments show that the proposed approach achieves comparable image enhancement performance and significantly improves the effectiveness of downstream vision tasks.