Sunday, 11 January 2015

Programming - Elite Dangerous Tools - OCR Pt 2

Update: You can now support this project at Patreon!

Some progress with the OCR I'm implementing, I've now broken the black and white filtered image down into rows, and then each row into the characters, and then I've looked and analysed where the gaps are between characters and decided where the columns are....

So here we see my pattern matching (simple pattern matching - not a neural network) reading through the characters and building the words for column 1, which is the name of the commodity...

As you can see it is working quite well... I need to keep training the pattern matching weights more, and this is very memory intensive (it takes over 300MB of data to process the various images, sub images, patterns and pixel values I'm summing & weighing against one another).

The biggest stumbling blocks is that the sourcing of the images from the client is very user dependent and twitchy, there are rules for it and ways for the user to capture the image in order and with the best overlap to help the stitching algorithms.

But this is the first pass of this tool, this is for me, to help capture and use market data personally... When I've collected a bunch of such data I may see if there are interested parties in helping collect more and more data.

Until then however, keep yourselves informed by popping back to the blog from time to time, and as ever let me know if you would use a tool like this!  Or even sponsor me to develop it further!

No comments:

Post a Comment