Deep learning: visual keyword spotting in historical handwritten documents

This is a deep learning based computer vision project.
The objective is to create a visual word seach engine so as to enable historians to search efficiently through the Ottoman Document Archive.
The input will be an image of a word. And the output is expected to be a ranked list of documents containing that word.
It involves similarity learning & metric learning, and multi-input architectures.
Students with prior experience about deep learning will be prioritized.