In this Article
A lot of time has passed in the meantime. The year is 1530 and Leonardo has been dead for 11 years. Many designs for war machines lie dormant in his legacy. Michelangelo Buonarroti is asked by the leaders of his home city of Florence to equip the city with machines for its defense. The emperor and the pope are standing outside the city with their armies. Michelangelo draws on Leonardo’s drawings and then has war machines built.
Is he allowed to do that? One thing is certain: he wants to change the logos, names and ownership notices in all the drawings. And there are several books full. But SEAL Systems can help.
The task
This is why an automated process is needed …
- that identifies and removes the documents in document management
- which in the meantime marks each data record in progress in the DMS accordingly
- to find the position of logos, names, copyrights, etc.
- that scales the new information and stamps it in the right place
- optionally allows interactive quality assurance on a random sample basis …
- and stores the changed files back in the DMS.
Logos and company names are usually found in the title block, but their exact position is unknown. The human eye sees them immediately. But how do you automate the finding process using software?
Seek and ye shall find
A characteristic on the relevant document info records indicates that an original has been added to the adjustment process. This allows other processes to react accordingly when accessing such an original.
The software is provided with a number of templates for training, which support the search.
The DPF-based workflow uses OpenCV (Library for Computer Vision) and possibly supported by AI methods (Tensorflow or Azure AI Services) to determine the enclosing rectangle for the logo, company name, copyright and confidentiality notice.
If the items searched for are not found, the file in question is sorted out as incorrect. It can now be reprocessed manually or used to train the recognition again.
The software also determines the trustworthiness of the position found. Hits with low confidentiality can therefore be checked manually to avoid errors.
To ensure quality, a certain quantity of conversions must also be viewed by humans. The system automatically removes a predefined subset at random. The operator is then offered the checklist in a separate client with a preview option.
The overall process looks like this:
Summary
A project like this also shows that there are ideas for the use of AI (artificial intelligence) in the field of Enterprise Output Management. It is worth looking into this and thinking about where AI can be used to improve or even automate processes in the company or at the customer.