Recognize in NI Vision assistant

[Johnson1213]

I want to crop the license plate, learn and recognize it.At this step, I don't know what to do next. Can you give me some suggestion>

[RamonG]

Can you tell us a little more of what you are trying to achieve, like do you want to match this exact plate in multiple images (pattern matching/geometric matching)?

If so you won't need a binary pattern but a grayscale pattern, preferably with little perspective distortion.

[Johnson1213]

I just want to find the license plate on multiple images, like finding rectangles in different license plate images.

[RamonG]

If you're looking for the same license plate in multiple images you could go for pattern matching (or perhaps geometric matching) and create a decent rectangular pattern. Use the NI example finder and search for "matching" to get you started.

Finding any license plate in multiple images and reading its characters requires a whole new (more complex) strategy. The Vision assistant can help with coming up with some pre-processing and setting up OCR to learn the characters, but ultimately you would need to save your script as LabVIEW code for more flexibility and take it from there.

My strategy would be something like this:

• Use color threshold to find the white part of the license plate
• Fill the holes to make it a rectangular shape  
• Use shape matching to look for rectangles, this outputs an array of ROIs  
• Use extract tetragon by feeding the ROI to remove any distortion  
• Use Identification - OCR/OCV in Vision assistant to train font recognition, you might want to use a gaussian blur for better separation

 I hope this helps to get started.

[Johnson1213]

I try to do it. I can not to solve it. Help

[Johnson1213]

i find the license plates Number in to Visual Assistant, but i can not to extract it to other place,and it has some noise.  I can't go to the next step which OCR recognition.

[RamonG]

See example below.

It seems you can also detect shapes (i.e. rectangles) in grayscale images. You will need to tweak the shape detection settings to your liking. This will give you the contour points.

I don't see the Extract Tetragon in Vision Assistant so exporting to LabVIEW could give you something like this:

 

You can save the extracted rectangle as a new file and feed it back into Vision Assistant for the OCR training.