This paper presents a new extracting method for several types of texts from a text/graphic mixed document image. We also propose a new word grouping method when intersected words are placed on a circular arc or any line segment with an arbitrary orientation. The basic strategy of our algorithm is based on the analysis of the run-length of the document image. The average and variance of the number of runs in a run-length encoding provide a nice structural property for symbols and texts. We propose 3-dimensional neighborhood graph for grouping word from a set of isolated characters, which are obtained from the first character-isolating phase. This graph maps each letter to a vertex in 3-dimensional space according to the ``volume'' of the character. Experimental results show that more than 97% of words were successfully extracted from a text/graphic mixed document.

keywords:document analysis, pattern recognition, text extraction, image processing