<p>Image quality enhancement remains a critical challenge in image processing techniques, especially for images captured under suboptimal lighting conditions that can generate low contrast, color distortion, and noise. These degradations not only affect visual aesthetics but also hinder the performance of downstream image processing tasks. The primary objective of image quality enhancement is to improve the visual quality of such images to benefit subsequent processing. Despite extensive research, achieving high-quality enhanced images remains challenging. Traditional image quality enhancement techniques often address only overexposure or underexposure, which can fail when both issues are present. Deep learning has recently been increasingly adopted in image processing, demonstrating significant potential to enhance image quality with underexposure, overexposure, or a combination of both. In this paper, we enhance the image quality over prior work by incorporating a convolution-based efficient transformer (CET). The proposed approach consists of four main stages. First, we enrich the training data by building a new dataset. Second, the collected dataset is preprocessed to eliminate noisy and low-quality samples. Third, CET is integrated into the Color Shift Estimation and Correction (CSEC) architecture to enable more effective feature extraction. Fourth, an additional depth-based loss function is introduced, leveraging depth maps to improve accuracy and consistency in image correction. Finally, the enhanced CSEC model is trained on both public datasets and the newly constructed dataset. The experimental results demonstrate the effectiveness and superior performance of the proposed approach compared to existing methods.</p>