HWR TEXT SEGMENTATION BASED ON SUBBAND ANALYSIS

Author(s):  E.G. Zhylyakov, Dr., Prof., Belgorod National Research University, Belgorod, Russia, zhilyakov@bsu.edu.ru

A.A. Chernomorets, Dr., associate Professor, Belgorod State National Research University, Belgorod, Russia, chernomorets@bsu.edu.ru

N.O. Efimov, Belgorod National Research University, Belgorod, Russia, chernomorets@bsu.edu.ru

Issue:  Volume 46, № 1

Rubric:  Infocommunication technologies

Annotation:  In the article we introduced method of word handwritten text segmentation. One of the major problems for successful text recognition (in the sense of small probability of errors of the first and second type) is the complexity (sometimes even the impossibility) of segmentation handwritten text into characters. At the same time, handwriting images have the periodicity property- for same author, the main characteristics of handwriting, the size and character of writing symbols, words and lines are preserved. For analyze quasi-periodic and periodic details on the image, adequately use frequency representations. Analysis of the frequency space from the standpoint of partitioning into subintervals is called subband analysis. Using subband mathematical apparatus allows to separate background areas and the areas of text for given image. Using accurate image energy parts values in the given subareas of spatial frequencies as weight factor allows to accurate localize words, small objects and artifacts for noisy images. The efficiency of the method is confirmed by presented results of computational experiments conducted in the Matlab software package.

Keywords:  handwritten text, image processing, image segmentation, handwritten text segmentation, HWR, subband analysis

Full text (PDF):  Download

Downloads count:  274