CODEWITHSUNDEEP
image-processingmachine-learningcomputer-visionocr
Ibrahim Amer
Ibrahim Amer

Reputation: 1218

Text detection feature extraction from text document images

I want to develop an application to detect text from scanned documents which may contain mixed text, graph, images, etc .. I think I can use sliding window technique of fixed size and using a feedforward NN or SVM to classify these regions, but I'm stuck with feature extraction phase what are common used features extraction methods to use ?

Upvotes: 2

Views: 2811

Answers (2)

ikkjo
ikkjo

Reputation: 785

The following might help you. Perhaps it is overkill for what you are trying to do, but it is meant to identify text regions in an image.

"Robust Text Detection in Natural Images with Edge-enhanced Maximally Stable Extremal Regions" (http://web.stanford.edu/~hchen2/papers/ICIP2011_RobustTextDetection.pdf)

You can find a related project on github (I have not used this myself): https://github.com/subokita/Robust-Text-Detection but it might be worth a try with your example images.

There are a lot of scientific papers about page/text segmentation, search for those terms on google scholar and you can find older, less complex approaches. One example is "Page segmentation and classification using fast feature extraction and connectivity analysis" (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.55.527&rep=rep1&type=pdf)

Upvotes: 1

Heshan Sandeepa
Heshan Sandeepa

Reputation: 3688

If all the characters are machine printed, one of easy and accurate feature would be the normalized density value of each pre-processed character image.

There you can divide the image in to fixed number of segments and calculate the normalized density value for each segment. Then you will get vector of density values for each character and proceed with the classification with NN or SVM.

Upvotes: 1

Related Questions