With the rapid development of digital technology, graphics are playing an increasingly important role in many fields, such as medical imaging, satellite remote sensing, and security monitoring. However, due to factors such as equipment and transmission, graphics often have poor clarity such as blur and distortion, which limits their effective application in practical scenarios. This paper systematically deblurs and enhances graphics by constructing a targeted convolutional neural network model to improve clarity. First, the classic convolutional neural network architecture is optimized to design a network structure suitable for optimizing image clarity. Second, transfer learning and data enhancement techniques are used to improve the model training effect. The experimental results show that the PSNR is improved by 2.64 dB, the SSIM is improved by 0.047, and the VIF is increased by 0.084 compared with SRCNN, proving that the model is effective in improving image clarity and structural similarity. However, due to the increase in model complexity, the inference time is also prolonged accordingly. Compared with traditional methods, the proposed convolutional neural network model significantly improves the clarity index of graphics in various graphics clarity optimization tasks and effectively improves the visual quality of graphics.