Accurate OCR Engine

We are looking for someone to create a library that can extract information from saved image files (in .PNG format). The saved images will be scans/photographs of driver's licenses, state IDs, federal (USA) IDs, and optionally other international IDs. ## Deliverables The library needs to return a value indicating if the information (Name, address, ID #, etc.) is guaranteed accurate, or if there is a problem interpreting some of the data. If there is a problem, the best-guess information should be returned. The library needs to return guaranteed information 99% of the time. So, if 100 IDs are scanned, the text returned for 99 of them must be perfect. For 1000 IDs, 990 need to be perfect. Your code should be able to identify approximately 20 types of IDs. Several types are only slight variants of another type and should require minimal extra effort once the initial type is able to be recognized. I WILL provide actual images for each type of ID to be identified. There will be multiple samples for each type of ID. Upon starting the project, the winning bidder will be provided with at least several hundred total samples. The ultimate goal is for the program to work with IDs from all 50+ US states and a select number of federal IDs. If you do good work, we will likely ask you to add to this in the future. The information required from each ID is as follows: * Issuing authority (State/country) * DL/ID number * Form (Driver's License, CDL, ID, Under 21 Driver's License, Under 21 ID, Temporary, etc.) * Type (Original, Duplicate, Correction) * Last Name * First Name * Full Name (everything on the ID; first, middle, last, suffix, etc.) * Address1 (Typically number and street) * Address2 (Rare; usually blank) * Birth Date * Issue Date * Expiration Date * Sex (M/F) * Height * Weight * Eye Color If a particular ID does not contain one of the above fields, the library should return a blank string for that field. The zip file attached to this post shows two sample scans of a credit card. The scanner used for these samples is the same one that is used to scan the actual IDs.