License Plate recognition using AForge.Net

Question

i am currently doing a project for "Automatic Number Plate Detection and Recognition" and i am using AForge.Net ;i have used a simple algo; though it is not very efficient but it will fulfills my requirements. i have calculated height,width and height to width(height/width) ratio of more than 40 car's numberplate; i searched for the rectangles in the image which lie in the range of height, width and height to width ratio; now with this i am able to detect the number plate for most of the images but the problem is that along with the number plate; i am getting some other rectangles; which also lies in the range of height/width ratio; now i want to differentiate between the rectangle containing the number plate and the rectangles without the number plate; [actually i am cropping the rectangles and storing them in Bitmaps] now i have 2 solutions to solve this but i don't know how to implement them.

i. i can make an assumption that the background of number plate is white(it is not true always but it will work for most of the cases (i am ready for the trafeoffs)); my question is that is there any built-in method in Aforge.Net that can search the white background from a picture; or how can i find that the given picture has white background?? [if a picture has a white background then it is a number plate else not]

ii. another option is to detect the characters from the image; any image containing characters [we can assume that there should be atleast 4 character; so we should search in an image that if it has 4 or more than 4 characters then it is a number plate else not] ; so is there any method available in AForge which can be helpful in this case i.e which can search for characters; if not then how can i search the characters through an image...?

Is there any other options; other than the two above ??

If for above 2 implementations; there is a solution in emguCV , even then do let me know; i can do a portion of work with emguCV also.

PLz reply me early as this is my college project ; and now only a little time is left...

regards usama

Answer 1

Turn the rhonmbus into a rectangle, divide the rectangle into 4x8. Compute median luminence in each rectangle and curve fit to known data set.

Run a guassian over the image before computing the normalized luminance to account for stickers and scratches and camera anomalies. Desaturate and quantize into 4 levels or so, the lightest color in the picture will be 0 the darkest is 100.