deep-anpr Implementation on Natural Scene Text

Hi all, I am trying to implement this work for text detection from natural scene images. Has anybody implemented it for that?

Thanks.

Apr 14 '17 01:04 guddulrk

I have played around with this code a bit, but I see a limitation in this work when it comes to detecting different patterns of text and numbers or variable lengths . The code was written to train the network to recognize a fixed pattern of a fixed length. That said, it does well on detecting that pattern on any background, but as for your case I do not know if this will work.

Apr 14 '17 07:04 sourabh2k15

i can give a generic answer to why this is happening , I think it is because the sliding Window used here is not trained to do so. We could add a few more layers and make this more extensive and train the network to detect characters better . Also the variable length issue can be solved if we can first have a binary classifier for text and non text and do some smoothing on the text region and have our network process that part . But that would mean a Major Change in the network architecture . Anyone up for it ?

Jun 23 '17 11:06 orgh0

Would love to help :+1: , just provide me some pointers to learn more about this stuff, bit of a newbie in this stuff.

Jun 23 '17 14:06 sourabh2k15