SUGG: Font Recognition and Font Similarity Learning

[ghost]

As the development of OCR is moving towards LSTM, new difficulties arise, such as having created lstm/traineddata for different fonts.

Example situations:

• I have created multiple lstm/ traineddata models, each for a different font and form of writing, and want to choose the most suitable one.
• Or, I just want to identify the font type from an image, and similar font types, instead of a full ocr.

I suggest integrating a new feature which gives the ability to train deconvolutional networks in identifying the Font Type and provide similar font matches suggestion from an image.

Demostration: An example of such implementation would be the Adobe DeepFont, have a look:

• Video of Deep Font demonstration

• Document

• Detailed research

• Python script demonstration

• The AdobeVFR_release files

• The developer page, he suggests using cuda-convnet as well as anna package for training deconvolutional networks.

• Whatthefont

Related research: A Sequence Learning Approach for Multiple Script Identification

The result: I will know what is the name of the font from an image, along with similar font matches to it.

Thank you all for your contributions: @theraysmith @tmbdev @mittagessen @zuphilip @zdenop @stweil @amitdo @Shreeshrii @kba @zuphilip