OCR receipt module for app (iOS/Andorid) - Proof of concept

General information for the app: this is a module function required as part of an app
Description of requirements/features: I need to develop functionality for an app to scan receipts and perform Optical Character Recognition (OCR) on them to extract certain fields value like business name, line item, prices, taxes (VAT), total etc within a response time of 3-5 seconds with accuracy rates of 99%. Web application should enable machine-learning to gradually improve the OCR based on the corrections made by the users.
The solution must use on-device resources for OCR, not server side and be able to handle following:

1) Noise and quality of the receipt/bills and be able to extract data no matter what the type of conditions be it lighting, paper print quality etc.

2) For iOS and Android, adjust manually for brightness and rotation.

3) Automatic image-optimizing features. For example, compensate for the skewed perspective that occurs when you don't manage to hold the iPhone camera precisely parallel to the document you're shooting, or when the image from a receipt is slightly curved/folded.

4) Sense where the edge of the document is, to ensure that you don't take a picture of a page and somehow leave out some of the text.

5) Automatically adjusts the perspective (portrait/landscape), sharpness and contrast of the camera to optimize images of text to create a sharper image of the text. It should also automatically straighten skewed or off-center images.

This proof of concept is to be able to read/extract these items

Please let us know how you intend to accomplish this job and what OCR platform / API/ libraries you plan to use. Expertise with image enhancements technologies might be of some help.
Similar urls: ocrapiservice.com - although this is server side
Extra notes: I have seen apps that are supposed to do this but they all seemed not to work!