Reference:
[1]lenet. http://yann.lecun.com/exdb/lenet/
[2] Ian J. Goodfellow, Yaroslav Bulatov, Julian Ibarz, Sacha Arnoud, Vinay Shet. Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks. http://arxiv.org/abs/1312.6082v2
[3] Sahu, Devendra Kumar; Sukhwani, Mohak.Target-Dependent Sentiment Classification with Long Short Term Memory. http://arxiv.org/abs/1511.04176
可以参考一下HCCR-GoolgeNet,虽然这是一篇关于中文手写字体识别的文章,但是在关于怎样使用现有模型上有一些启发性
参考文献:
High Performance Offline Handwritten Chinese Character Recognition Using GoogLeNet and Feature Maps
源码:
github: https://github.com/zhongzhuoyao/HCCR-GoogLeNet
3 个回复
betas - 半路出家
赞同来自: xinmiao 、OpenAI
如果能够分割好图,对分割的图,用一般图片识别的CNN模型就行了,网络的深度也不用很深。
如果你不想分割图,直接end-to-end,可以看看google街景识别门牌号算法[2],最近貌似用RNN的paper比较多,而且我之前面过某家做OCR的公司,面试官说他们最近尝试用lstm[3],效果还不错。
Reference:
[1]lenet. http://yann.lecun.com/exdb/lenet/
[2] Ian J. Goodfellow, Yaroslav Bulatov, Julian Ibarz, Sacha Arnoud, Vinay Shet. Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks. http://arxiv.org/abs/1312.6082v2
[3] Sahu, Devendra Kumar; Sukhwani, Mohak.Target-Dependent Sentiment Classification with Long Short Term Memory. http://arxiv.org/abs/1511.04176
希望我的回答对你有帮助。
ps.
不是做这个的,有搞这个可以review一下这个答案。
xinmiao
赞同来自: OpenAI
参考文献:
High Performance Offline Handwritten Chinese Character Recognition Using GoogLeNet and Feature Maps
源码:
github: https://github.com/zhongzhuoyao/HCCR-GoogLeNet
shuokay
赞同来自: