How can i make the bounding box scan an image for characters?

[amt1981]

Dear All,

I am creating a License Plate recognition using LabVIEW 8.6, and i defined my script in LabVIEW Vision Assistant, now my question is how can i make the bounding box that we need to use in the OCR Session move automatically around a masked image to search for characters in a license plate?

Now i know that first i need to create a coordinate system for my masked region of the original image where most likely the License plate is found, and here also i am having a problem in doing so(Setting a coordinate system)!! So please if any one can help me do so.

N.B: I am not an expert in LabVIEW i recently started to learn to use LabVIEW so i am kinda new!

So any help would be appreciated!!

Thanks in advance

[muks]

Please post a few images of varying position of the characters. Is the background uniform and really contrast different? Can the entire characters be given as a template for patternmatching and get the bounding box from the result?

[amt1981]

Thanks for replying,

do you mean images of the trained characters or the image of the car itself? As for the difference in contrast i changed the contrast, brightness, and gamma properties of the image in order to enhance the number plate, and then i am feeding the new image into the OCR seesion. as for the bounding box, i am drawing it in the OCR step around the characters of the license plate and then Vision Assistant(OCR) is spliting the characters and then recognizing them! So i need to make the step where the bounding box will perform an automatic scan of the the masked region of my original image to find the license a region where he can recognize characters which will be the license plate.

[muks]

I meant the images that you are working with...

[amt1981]

Ok these are some of the Images i am working on. They are of 2 megapixel resolution.

[muks]

And so I want to tell you my views. IT is definetly not easy. First the challenge is to extract the plate alone and from that with all those variable illumination or shall i say ambience we have to train the characters. well lets work on it and try..

[amt1981]

Well actually i did a character training set for numbers ranging from 0 to 9 in many positions to let it recognize as much as possible all the characters in a License Plate.

And so far it is doing the job!! My problem is to automate the procedure of character reading when we get to the OCR session. Meaning i need the bouding box that will be surrounding the characters of the number plate to scan by itself the entire masked region of the  acquired image and locate the license plate and read the characters. In other words whenever it detects a known character it starts to scan for more characters at the same level (or y-axis position and varying the x coordinates) for a character set of 4 to 7 characters. 

Also it could be done by making a very specific masked region of a limited set of coordinates (like making the masked region to be of a new resolution of 676x447 for example) and make it scan this region to locate the license plate and read its characters.

This should be done in a VI (which is generated from Vision Assistant) and in this VI i should manipulate it to scan automatically the area of the masked image to locate the license plate and read the characters. And this is where i am being confused because of my lack of knowledge in LabView 😞 LOL

So i thank you for your assistance. I will be waiting for your solution for the problem and meanwhile i will try to see how i can do this job!! 😄