No/VOL: 05/2013 Page no. 117
Authors: Matko Sarić , Hrvoje Dujmić , Mladen Russo :
Title: Scene Text Extraction in HSI Color Space using K-means Algorithm and Modified Cylindrical Distance
Abstract: Text extraction, that is segmentation of characters from background, is especially important step that greatly determines final recognition performance. Particular focus is put on this task for scene text which is characterized with wide set of degradations like complex backgrounds, uneven illumination, viewing angle etc. In this paper we introduce text extraction method based on k-means clustering with modified cylindrical distance in HSI color space. Performance of this distance is analyzed depending on different degrees of chroma reliability. For purpose of result comparison, K-means text extraction is also performed with cylindrical distance in HSI color space and Euclidean distance in RGB color space. Complementarity of tested distances is also analyzed showing possible direction for further performance improvement.
Key words: scene text, K-means, cylindrical distance.