Training

GiDoc is able to train HMMs and language models used for recognition.

Open an image and click ``GiDoc /Training''. With independance of the image from which the plugin has been called, the system will perform training using the transcriptions stored in the xcf image files located in the Document Directory option of the project preferences. If the folder does not exist, it will train using only the opened image.

Training takes place using examples, therefore at least some pages have to be transcribed previously. As more transcriptions are available, better models are obtained. Hence, Training must be called every few pages have have been transcribed. This way the recognition system's performance is improved. Concretely two options are avaible in this process, to train from the begining all the models or to restimate the current ones.