Recognition of low-resolution characters by a generative learning method
Hiroyuki Ishida, Shinsuke Yanadume, Tomokazu Takahashi, Ichiro Ide, Yoshito Mekada · 2005
Using appropriate training data is necessary to ro-bustly recognize low-resolution characters by the subspace method. Former learning methods used characters actu-ally captured by a camera, which required the collection of characters of all categories in various conditions. In this paper, we propose a new learning method that generates training data by a point spread function estimated before-hand by captured images. This method is efficient, since it eliminates collection of the training data. We confirmed its usefulness by experiments. 1.