An Application Combining CNN-BLSTM with CTC for License Plate Recognition

EasyChair Preprint no. 7241

Abstract

This paper proposes an application based on convolutional neural network-bidirectional long short term memory and connectionist temporal classification for the problem of license plate recognition. Unlike previous machine learning models, the hybrid network handles variable-length sequences and does not need to segment characters from plates. The application experiments with a model of hybrid network, evaluates its correct predictions on the validation set by BLEU scores, and compares it with the model of K-nearest neighbors and the model of Support vector machines. Estimated BLEU scores show that the model of hybrid network gives a reliable accuracy, and this result is better than that of two machine learning models.

Keyphrases: beam, connectionist, Convolutional, Recurrent

@Booklet{EasyChair:7241,
  author = {Pham Tuan Dat},
  title = {An Application Combining CNN-BLSTM with CTC for License Plate Recognition},
  howpublished = {EasyChair Preprint no. 7241},

  year = {EasyChair, 2021}}