Several OCR tools on printed text is available, also a few HCR tools/apps are available that aims to convert handwritten text into digital form. But I have some documents that contains both - certain portion is printed and rest are handwritten. Please suggest a system that provide good accuracy in such mixed text.
For printed text you can use Tesseact in Python and for Handwritten you will need to train your machine for Extended-MNIST dataset or find some other handwriting dataset. Then after training detect/segment your handwriting and then you can recognise.
For more information you can email me at: [email protected]
Hi, if you are working on text recognition of a language (script). So, you can train the network who will be able to test printed and handwritten text. Means, your dataset will be the combination of the printed and handwritten text. then you will use that mixed dataset to train and test the network. You can go with supervised and unsupervised learning but it will depend on your literature.
If you are working on recognition of famous language (English). So, there are per-trained networks (VGG) are also available then you can use that network.
For more details, can contact me anytime.
thanks
Hello, For printed text only you can use Fuzzy logic system and for Handwritten only you can train supervised neural network. if you use dataset including combination of the printed and handwritten text I think may be supervised neural network is more accuracy in such mixed text.
Sorry, but for my understanding, if you want to go with supervised learning so, feature extraction is most important part and ANN, SVM can use to classify them. In current time, unsupervised leaning is performing very well but it require a big dataset compare to supervised learning.
So, you can not say that a technique perform well to recognize the characters. Several things have to focused to find out a suitable technique to reach maximum accuracy.
As per my knowledge i can say these things, if you have document images you have extract the text part from that. then do some preprocess on them like removing noises, extract the potential features from the preprocessed images. store feature vector with corrosponding labels( here labes are those which can related to particular class) then you can use classifier there are many classifiers are available i.e., KNN, SVM, LDA and so on. for the classification you will have app called classification learner app in Matlab 2015b onwards versions. By using classifier you obtain the recognition accuracy result. if the accuracy is nearer to the 100% then that result is treated as a highest result.
with Best wishes,
SATISH