Enhancement of Blurred Indonesian License Plate Number Identification Using Multi-Scale Information CNN

One of the challenges in traffic surveillance and public safety in Indonesia is the recognition of blurred vehicle license plates. In this case, the development of image processing technology and artificial intelligence is key in improving the ability of license plate recognition systems to overcome the obstacles of blurred images. This research aims to develop a license plate recognition method that can work effectively in overcoming the problem of blurred license plate recognition and improve accuracy. This research will use advanced and accurate image processing technique technology by utilizing multiple layers, such as Convolutional Neural Network (CNN), Multi-scale Information CNN (I-CNN), Number Plate Segmentation, and Transfer learning. In addition, the proposed number plate recognition method is able to cope with blurred images and variations in the characteristics of number plates in the form of different letters and numbers throughout Indonesia. In this study, we used the License Plate Digits Classification (LPDC) dataset consisting of 35-character classes. Each class consists of 1000–1030 photos from different angles and lighting conditions. The dataset we tested was an Indonesian license plate, consisting of 549 blur datasets and 255 normal datasets. By using categorical cross entropy to determine the accuracy in this research, we get the results of CNN training accuracy of 85,73% and I-CNN training accuracy of 97,69%.