The Implementation of Real-ESRGAN as An Anticipation to Reduce CER Value in Plate Number Extraction Results Employing EasyOCR

The accuracy of the number plate extraction is an important thing that must be considered. This was done to prevent false positive number plate extraction results in the e-ticket system. The efforts to reduce the most optimal CER value could be done by upscaling utilizing Real-ESRGAN, especially by operating anime mode which had been done with grayscaling and thresholding processes. This treatment was able to reduce the average CER and CER Norm values, respectively, at 30.97% and 29.72%. When compared with the treatment before the image was upscaled, namely Original-Skewed (Grayscaling + Thresholding), this treatment had a lower average CER and CER Norm about 10%. Anime mode in the upscale process utilizing Real-ESRGAN without grayscaling and thresholding processes were able to produce pixel arrays that have minimal noise or excess texture as well as emphasize outlines. This may reduce the number of image features that cause an increase in the CER value but make the thresholding process cleaner because the noise in the image is much reduced.