Text document photos remain a highly important resource in the development or training of
document segmentation and Optical Character Recognition (OCR) applications. This is due to the
quality assessment of algorithms for these tasks is performed using unknown images, which
guarantees their robustness, and that Machine Learning techniques need large datasets with a great
number of samples with high inter and intraclass diversity. This work presents a method to
procedurally generate simulations of photos of text documents with different backgrounds. Using
Digital Image Processing techniques, this process eliminates the need of creating a real dataset by
hand and increases the variability of the available data. The dataset generated can have any number
of samples, can be used in models that need smaller or larger datasets.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
INC, HP, "PROCEDURALLY GENERATING ARTIFICIAL PHOTOS OF TEXT DOCUMENTS IN VARIOUS BACKGROUNDS", Technical Disclosure Commons, (December 09, 2020)