7
$\begingroup$

I am trying to build and optical character recognition system for recognizing license plate (Indonesian licence plat), unfortunately there is no training set available but I found the font, I try to generate the training data by convolve the image of license plat letter with kernels (somethings like gaussian blur,box blur) using python, but it does not resemble real data, here is the font look like: enter image description here

and I want to generate individual letter, that look like this:

plat kendaraan

Do you have any idea how to generate training data like above picture, ? thank you

$\endgroup$
  • $\begingroup$ What about creating a dataset? What is your system used for? Can you get unlabeled data easily? $\endgroup$ – Martin Thoma Nov 29 '16 at 22:11
  • $\begingroup$ It is dificult to get dataset, even unlabeled data set, the dataset should be a croped image of car plate license letter, the system is used for reading license plat number $\endgroup$ – Kiki Rizki Arpiandi Nov 30 '16 at 2:47
  • $\begingroup$ I used deep convolutional neural network for classifiying digits, stroke width transform for detecting the text and k-means clustering text segmentation $\endgroup$ – Kiki Rizki Arpiandi Nov 30 '16 at 2:51
7
$\begingroup$

That's a very good question. I am facing a similar issue with different data, but my research led me to this: https://matthewearl.github.io/2016/05/06/cnn-anpr/ which should be pretty much what you are looking for.

$\endgroup$
  • $\begingroup$ Wow, thankyou, that is pretty cool, $\endgroup$ – Kiki Rizki Arpiandi Nov 29 '16 at 0:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.