car plate recognition
-
any idea how to extract car plate number from jpeg image?? i trying to write this code but i need some assistance..... mind to share with me...a source code would be nice....:sigh:
Well, most car plates are done in a standard font If you could create an array of images / vectors showing the framework for each possible character (ie the skeleton of an F would be just the basic lines), and then had it search for this shape within each letter in the license plate. What you have to allow for are lighting, the angle at which the photo was taken, possible different sizes, speeds of car etc. Looking for a framework will allow you a pretty accurate measurement, and you can make countermeasures against specific problems (ie images on license plates etc) Regards James
-
any idea how to extract car plate number from jpeg image?? i trying to write this code but i need some assistance..... mind to share with me...a source code would be nice....:sigh:
Hello jianloong, If you find out car plate then i send you link of some article. Which contain source code with example demo. http://www.codeproject.com/cs/media/Motion_Detection.asp[^] http://www.codeproject.com/csharp/motion_detection.asp[^] http://www.codeproject.com/audio/featuretracking.asp[^] After detection of car plate you can use OCM(Optical character reader) software tool. Actually at this portion i can't help you more. Best of luck. Regards, Divyang Mithaiwala
-
Hello jianloong, If you find out car plate then i send you link of some article. Which contain source code with example demo. http://www.codeproject.com/cs/media/Motion_Detection.asp[^] http://www.codeproject.com/csharp/motion_detection.asp[^] http://www.codeproject.com/audio/featuretracking.asp[^] After detection of car plate you can use OCM(Optical character reader) software tool. Actually at this portion i can't help you more. Best of luck. Regards, Divyang Mithaiwala
-
thats great guyys...thx a lot for the character recognition, i got think about neural network...i know its the best solution but its hard to write for me knowing that my programming level is not that good.. any suggestion of other ways to do it??
I've programmed a simple neural network in some hardware before and I've simulated it in C++ too. You only need to first demystify the neural network by realizing it's just standard matrix algebra then figure out how to apply the neural network to the problem. I figure you'd have to create an image for each character on a place, feed each into the network so the network "learns" all the images, then scan the plate and pick out each letter. It's best to convert your pictures to black and white because the neural network works best when there's ample contrast. If you can't convert the pictures, then you'll have to increase the qualify of the network by adding new nodes. If you demystify and understand a neural network, the short paragraph above makes much more sense. Good luck. I know two groups of people that used a neural network to correctly identify someone by their voice and by their faces over 99% of the time.
-
I've programmed a simple neural network in some hardware before and I've simulated it in C++ too. You only need to first demystify the neural network by realizing it's just standard matrix algebra then figure out how to apply the neural network to the problem. I figure you'd have to create an image for each character on a place, feed each into the network so the network "learns" all the images, then scan the plate and pick out each letter. It's best to convert your pictures to black and white because the neural network works best when there's ample contrast. If you can't convert the pictures, then you'll have to increase the qualify of the network by adding new nodes. If you demystify and understand a neural network, the short paragraph above makes much more sense. Good luck. I know two groups of people that used a neural network to correctly identify someone by their voice and by their faces over 99% of the time.
In response to the latest reply: Im not too familiar with neural networks, but from what ive heard, am I right in thinking its similar to reverse-engineering matrices in DirectX to find what object a mouse clicked in a 3D world? if you are trying to simulate a speed camera (i.e. a stationary object) then you may not need this as you can use the same algorithm as the cars will always be taken in the same position (roughly), and you can search the whole scene for a license plate (look for yellow etc) and then scan that area for something which has the skeleton of an "A" and so on. ie, if you get any font on MS word, (well most sensible ones) you can always fit a skeleton into each letter - they should be one pixel thick, and fit into them. If you understand that, well done, if not its my explanatory skills lol - jamespraveen@aol.com MSN - that way I can explain in real time which may not be as bad. James (dont tell ANYONE im on AOL)
-
In response to the latest reply: Im not too familiar with neural networks, but from what ive heard, am I right in thinking its similar to reverse-engineering matrices in DirectX to find what object a mouse clicked in a 3D world? if you are trying to simulate a speed camera (i.e. a stationary object) then you may not need this as you can use the same algorithm as the cars will always be taken in the same position (roughly), and you can search the whole scene for a license plate (look for yellow etc) and then scan that area for something which has the skeleton of an "A" and so on. ie, if you get any font on MS word, (well most sensible ones) you can always fit a skeleton into each letter - they should be one pixel thick, and fit into them. If you understand that, well done, if not its my explanatory skills lol - jamespraveen@aol.com MSN - that way I can explain in real time which may not be as bad. James (dont tell ANYONE im on AOL)
I don't know about DirectX but I figure in DirectX that it simply converts a click and extends a vector basically perpendicular into the screen and the first surface it touches is the selected entity. If that's so, then I'm pretty sure it's not a neural network. You can use Windows fonts I guess but if you're reading license plate letters, you really should teach the network using license plate letters, otherwise your network might think 5's are S's and vice versa if a license plate S or 5 looks like Ariel's 5 or S, respectively. You want to "teach" the network with the best materials to minimize error when it goes to work. That's why I said objects that yield the highest contrast give the best weights to the network and do really well. Weights by the way are individual numbers on how likely any particular node is "on". For instance, if you fed in all the license plate letters and every character occupies a particular pixel, the node that corresponds to that pixel will probably have a weight of 1, meaning that pixel will always be used. When you to go set the network to operate and you give it a picture where that pixel is not used, then the progression of the network will rule out all of it's learned possiblities and result in telling the user that it doesn't know what that image is which is what is supposed to happen.
-
I've programmed a simple neural network in some hardware before and I've simulated it in C++ too. You only need to first demystify the neural network by realizing it's just standard matrix algebra then figure out how to apply the neural network to the problem. I figure you'd have to create an image for each character on a place, feed each into the network so the network "learns" all the images, then scan the plate and pick out each letter. It's best to convert your pictures to black and white because the neural network works best when there's ample contrast. If you can't convert the pictures, then you'll have to increase the qualify of the network by adding new nodes. If you demystify and understand a neural network, the short paragraph above makes much more sense. Good luck. I know two groups of people that used a neural network to correctly identify someone by their voice and by their faces over 99% of the time.
ermm...LighthouseJ..is it possible to do this character recognition without using neural network?..bcos i really dont have any base about neural network and for the starting...i might start with edge detecting to detect the edge bcos car plate normally comes in standard size..rite? do you have any example about character recognition?? it might help me a lot...and anyone out there if you all have anything to share with me just send to my e-mail jian_loong520@yahoo.com or just add me in my MSN...., mave_loong@hotmail.com thx a lot...
-
ermm...LighthouseJ..is it possible to do this character recognition without using neural network?..bcos i really dont have any base about neural network and for the starting...i might start with edge detecting to detect the edge bcos car plate normally comes in standard size..rite? do you have any example about character recognition?? it might help me a lot...and anyone out there if you all have anything to share with me just send to my e-mail jian_loong520@yahoo.com or just add me in my MSN...., mave_loong@hotmail.com thx a lot...
I can't think of any other mechanism that would do the trick. I don't have any of my old programs but there are several articles on here that feature neural networks in different ways and a google search for neural network yielded many pages on it. I'm sorry I can't help you further but I haven't used a neural network in a long time.
-
any idea how to extract car plate number from jpeg image?? i trying to write this code but i need some assistance..... mind to share with me...a source code would be nice....:sigh:
:)