I have multiple invoices images which need to classify invoice types such as fright, utility, goods, etc. Is there any way to classify without OCR?
It is possible to classify invoice scans without passing through an OCR component if they are visually different (they demonstrate different visual features). On the other hand, if the invoices look very similar, then the classifier might not be very accurate.
Another challenge would be the number of images you need to train a deep network for image classification (you may start with pretrained models and only perform the finetuning of you do not have enough images). On the other hand, the combination of pretrained OCR models and NLP-based document classifier may not need that many samples for training (for this specific task).