Hello all,
I’m working on ALPR, for that I would like to use only CNN, most of the famous OCRs like PP-OCR and EasyOCR seems to use both CNN and LSTM, but I want to use only CNN to implement ALPR, So do anyone have suggestions to implement it?
Yeah, it’s possible to build an ALPR system using only CNNs. You’ll just need to handle character segmentation separately since you’re not using sequence models like LSTMs. A common approach is to first detect the license plate using a CNN-based object detector like YOLO or SSD, then segment each character and pass it individually through a CNN classifier (like a lightweight ResNet or custom CNN). It’s a bit more work compared to end-to-end OCR systems, but it gives you more control and keeps it strictly CNN-based.