Word segmentation in handwritten Chinese text image based on component clustering techniques

Segmentation of handwritten Chinese input into individual character is a crucial step in many connected handwriting recognition systems. In this paper, a new method is addressed to segment off-line handwritten Chinese text images. We first adopt the HMM method to produce the segmentation paths and apply two rules to reduce the redundant paths, then the left candidate paths dissect the text line into radicals or pseudo-radicals-components. In the second stage, we propose three new criteria -aspect ratio, gap ratio, longer edge criteria - to calculate the clustering cost matrix and use a dynamic programming technique to produce the optimal clustering scheme. A series of experiments show that our method is very effective for the word segmentation of the offline handwritten Chinese text image.