Ask Your Question

Image Recognition question

asked 2013-04-17 22:57:27 -0600

rodsnjr gravatar image

updated 2013-04-18 08:55:05 -0600

Hi. I am currently working on a research project for mobile devices. The project is to create an app that must recognize if the image taken from the camera is a "dollar bill", after that the software needs to know what is the value of that note.

I think that should be done by comparing among other image mats stored on some file, or database and see what is the matching one. I've read a lot of papers and examples, most of them point that in some ways the image needs to be processed applying things like CannyEdge detection and other stuff. But now im not really sure which ones to use, and about the topic that recognition task could be classified on.

First Question: Which ones are the recommended Image Processing methods that i should use?

Second Question: What materials/topics related should i look to help me on the recognition part?

Second Question: What OpenCV(besides the initial part of Image Processing) have to auxiliate the task to classify if the image is a "dollar bill" and what it its values.

By "dollar bill" i mean it is "money" but not really dollar. Here's a sample image of the notes that i need to classify, i was thinking on using something to check the colors. But im not sure if thats precise enough.

Thanks for the help.

I tried using this "" it worked as a simple solution but it is only for desktop so i had to discard it.

I found the similar MLP on OpenCV but i have no idea how to use it and i'm not sure if there's time for me to completely understand how it works and implement something based on it.

I forgot to say, but at this phase of the project we're working on a simple solution and we're not wanting to track the bill on the image or anything, just check if it is or not it'll be an advance.

edit retag flag offensive close merge delete



Will the dollar bill be crumpled ?

If not, then you could try some edge detection algorithm, see if you could fit a polylines to it, and see if it's 4 sided. Extract that ROI, maybe warp the perspective (?) a bit, and calculate the color histogram of it, and see if it's similar to your expected histogram.

sub_o gravatar imagesub_o ( 2013-04-18 02:56:29 -0600 )edit

What have you tried? Did you google it at least?

sammy gravatar imagesammy ( 2013-04-18 03:22:01 -0600 )edit

Yeah, and already tried a solution, i forgot to put that on the question. Now i edited it. So i could save a color histogram for the image and mark it as with a label or anything like that, and compare the marked one with the one in the photo? This would be a simple solution, i think there wont be precision on it but it's at least a start. I was looking into the SURF algorithm on here "", i think it's useful for finding some features on the image, so i could compare this tracked parts on the histogram in real time. But i have no idea how to pick up these tracked parts of the image. And i'm still having difficulties on understanding the histogram result structure, is there any "histogram for dummies" somewhere? Again thanks for the help!

rodsnjr gravatar imagerodsnjr ( 2013-04-18 09:03:41 -0600 )edit

-1 for not using a descriptive title.

SR gravatar imageSR ( 2013-06-19 17:22:47 -0600 )edit

3 answers

Sort by ยป oldest newest most voted

answered 2013-04-18 14:23:57 -0600

Guanta gravatar image

Actually this is a very typical detection task and often answered in this forum. The most answers given: Either train a cascade classifier, see and the various questions about it on this Q&A forum. Or alternatively, you can try a bag of words approach combined with a classifier, see e.g.

edit flag offensive delete link more

answered 2013-04-18 03:01:20 -0600

Immi gravatar image

I think, for mobile device is checking colors no solution, because of huge variance of luminance.

edit flag offensive delete link more

answered 2013-04-19 02:11:48 -0600

Use haar training through openCV.. Feed it with positive and negative images, so that it identifies the pattern..!

edit flag offensive delete link more

Question Tools


Asked: 2013-04-17 22:57:27 -0600

Seen: 3,589 times

Last updated: Apr 19 '13