Receipts: Computer Vision Analysis of Documents

Square

Using the digital camera on any mobile device, the consumer could easily take a high resolution image scan of any small document e.g. receipts.

Ensemble of AI, Machine Learning, Computer Vision, Image Processing,  and OCR algorithms Extract useful data from the image scan.

Highlighted Bouding-Boxes appear to identify and extract a full line.

Separate algorithm extracts bounding-box for the barcode.

Extractions are placed into a database for processing by other applications or human operator if need be.

Note the yellow highlighted numbers are the confidence levels for accuracy of the extractions converted into ASCII text.

 

Issues

Generally speaking placing the receipt on a flat well lit surface should suffice, but the algorithms are able to extract from lesser requirements

 

 

Other objects e.g. fingers may appear in the original image scan, nonetheless the algorithms manage the extraction quite well.

 

 

 

Machine Learning algorithm Nearest Neighbor used on the color space of the image scan to reduce the blue-ish tint. Generally mos discolorations and non-participating colors are filtered out.