Object Detection Based Handwriting Localization

We present an object detection based approach to localize handwritten regions from documents, which initially aims enhance the anonymization during data transmission. The concatenated fusion of original and preprocessed images containing both printed texts notes or signatures are fed into convolutional neural network, where bounding boxes learned detect handwriting. Afterwards, can be processed (e.g. replaced with redacted signatures) conceal personally identifiable information (PII). This processing pipeline on deep learning network Cascade R-CNN works at 10 fps a GPU inference, ensures enhanced minimal computational overheads. Furthermore, impressive generalizability has been empirically showcased: trained model English-dominant dataset well fictitious unseen invoices, even in Chinese. proposed is also expected facilitate other tasks such as handwriting recognition signature verification.