2024-03-29T12:31:33Z
https://soar-ir.repo.nii.ac.jp/oai
oai:soar-ir.repo.nii.ac.jp:00013243
2022-12-14T04:02:21Z
1221:1307
Extraction of characters on signboards in natural scene images by stump classifiers Proc.10th International Conference on Document Analysis and Recognition
MARUYAMA, Minoru
YAMAGUCHI, Takuma
We present a method to detect characters on signboardsin natural scene images. For many applications, both classifierwith small computational cost and the efficient featureset, which gives rise to accurate recognition are required.Texture based features are often used for target detection. Ithas been also shown that the shape of the intensity distributionis often useful for character extraction. The intensitydistribution in the character regions is often different fromthe unimodal distribution. We measure the discrepancy betweenthe observed region and the normal distribution byskewness and kurtosis. We use these statistics along withthe texture based features. Character regions in a naturalscene image are detected by using the linear combination ofstump classifiers, each of which sees only one component ofmultidimensional feature vector. Selection of a feature componentfor each stump and determination of coefficients oflinear combination are carried out by AdaBoost. We experimentallyshow the effectiveness of the proposed method.
Article
ICDAR2009: 1365-1369 (2009)
International Conference on Document Analysis and Recognition
2009
eng
conference paper
AM
http://hdl.handle.net/10091/12016
https://soar-ir.repo.nii.ac.jp/records/13243
https://doi.org/10.1109/ICDAR.2009.147
10.1109/ICDAR.2009.147
1365
1369
https://soar-ir.repo.nii.ac.jp/record/13243/files/ICDAR2009-PID896088.pdf
application/pdf
2.1 MB
2015-09-28