Video Recordings

In Image Text Recognition pipelines, usually some hypotheses  about the letters in the image are made first. In the next stage the  best hypotheses are selected based on some criteria of how letters in a text can follow each other. The presented work is an extension of work  done at the Centre for Machine Perception of the Czech Technical  University. It is an application of the structured prediction methods on  learning decoding the hypotheses based on the image processing based on  a rich set of both visual and language features.