I have worked on desktop application for scanning, Zonal OCR based Indexing ( as is done by a Kofax Capture and EMC Capitiva products) so, I can be helpfull in suggesting a well tailored solution and some improvements as per given requirements.
Since cost is an important factor so, I would suggest using Tesseract OCR engine which is free.
The ABBYY OCR SDK would be too costly since they charge per/cpu license and ongoing royalties. I have worked with IDRS, Atalasoft, Leadtools and ABBYY OCR SDKs and found IDRS best thought it cost the highest.
Provided that you want full page/Image OCR and the doing some text search based on some keywords so, Tesseract is well suited. With good quality images, its accuracy is around 90-95%.
*It does not support reading multilingual characterset.
I am comfortable with any of the mentioned OCR engine SDK. I would be glad to discuss it further.
Thanks
Needo