Identification of text on colored book and journal covers

An approach to automatic text location and identification of colored book and journal covers is proposed. To reduce the amount of small variations in color, a clustering algorithm is applied in a preprocessing step. Two methods have been developed for extracting text hypotheses. One is based on a top-down analysis using successive splitting of image regions. The other is a bottom-up region growing algorithm. The results of both methods are combined to robustly distinguish between text and non-text elements. Text elements are binarized using automatically extracted information about text color. The binarized text regions can be used as input for a conventional OCR module. Results are shown for parts of book and journal covers of different complexity. The proposed method is not restricted to cover pages, but can be applied to the extraction of text from other types of color images as well.