Text zone classification is a vital step in the dig-itization process, without which OCR systems perform poorly.Prior methods to document zone classification have relied on largesets of hand-crafted features for training zone classifiers. Suchfeatures are usually database-dependent, and their computationis time consuming. In this work we propose a novel method fortext zone classification that relies on the approach of unsupervisedfeature learning. Within our method, feature vectors of documentzones are automatically learned by patches extraction, encodingand pooling, where feature encoding is based on a codebookof visual words. The training phase of the text classifier takesinto consideration the unbalance between text zones and non-text zones of all types. The proposed method has been tested onpublicly available standard databases, and achieved competitiveor better results compared to state-of-the-art methods. Theresults show that our approach matches well the task of textclassification, and is robust to zone shapes, orientations and size.

EXPLOR_STEP=$WICRI_ROOT/Ticri/CIDE/explor/OcrV1/Data/Main/Exploration

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000018 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Main/Exploration/biblio.hfd -nk 000018 | SxmlIndent | more

{{Explor lien
   |wiki=    Ticri/CIDE
   |area=    OcrV1
   |flux=    Main
   |étape=   Exploration
   |type=    RBID
   |clé=     Hal:hal-01319899
   |texte=   Text Zone Classification using Unsupervised Feature Learning
}}