In the field of forensic image analysis, the identification of license plates at the scene of an incident goes through the subjective visual identification process of the investigator. However, noise, blur, and resolution degradation of the image appearing under insufficient illumination become obstacles to human visual identification. In this study, the degraded license plate image is improved using ESRGAN. The model structure was modified to match the aspect ratio of the license plate image data. In addition, high-intensity augmentation was applied considering that the number of the license plate was damaged to the extent that it was difficult to recognize with the naked eye. To restore such a complex image, an encoder-decoder structure is added to the existing ESRGAN structure. In forensic image analysis, since visual discrimination is performed rather than an automated recognition method, performance evaluation was performed by subjective visual discrimination. The recognition rate by number was 60% before image improvement, but 87% after improvement, showing a significant image improvement effect.