google vision ocr api and returned text

[don magnify]

hello...

for quite some time now i have been experimenting with the google vision OCR api to convert store receipts to text and for the most part it has been working rather decently. 

one large issue that seems to be a challenge to resolve is that when the api returns the text scanned off a receipt ,90%+ of the time the prices on the receipt - usually on the far right side of the piece of paper on the submitted image - do not align with the line item that they belong to. the prices are still being read correctly (for the most part) but they get returned most often on random lines which are not associated with the line of the item to which they belong. 

i have been exploring the option of associating/stitching the item lines and the prices lines via the coordinates of the text chunks that are included in the api returned result but that method hasn't  proved very effective given the fact that the pictures are not always containing a perfectly straight image of a receipt, varying orientations, etc...

writing this post here with the hope that somebody here has experienced a similar issue using the google vision ocr api and has successfully resolved it. any ideas or suggestions are welcomed.

thanks...

[Eduardo Ortiz Caraveo]

I think you might find this tutorial helpful for your case.