Can I use a custom extraction model to detect signatures from ID documents?

Question

Hello,

I attempted to train a custom extraction model to detect signatures from various types of identity documents, such as IDs, passports, and driver's licenses. I trained the model with 100 documents, but when I tested it, it did not work—even on documents from the training set.

Could this be because the model is not designed for this type of task? Are there alternative models that could be more suitable for signature detection?

Thank you!

Answer 1

Hi Dalla Verde, Giada

Please make sure you have selected "Custom neural model" while creating. Custom neural model uses deep learning to find desired region of interest.

You can also segregate documents into different categories like and create separate custom extraction models and compose them through compose model operation (A classifier model is needed in minimum along multiple custom extraction models)

Custom compose model will use best model based on document and analyze.

Passport

• European
• American.
• ......

Driving license

• European
• Asean
• ....

Please use high resolution images and use preprocessing techniques like script to use uniform extension or converting larger files to small file sizes for improving inference time.

Attached documentation on Custom compose documentation for reference.

Please don’t forget to Accept Answer and Yes for "was this answer helpful" wherever the information provided helps you, this can be beneficial to other community members.

Thank you.