To precisely segment Chinese characters in images and videos, we propose a novel recognition-based method in this paper. Our method consists of four steps: 1) Original text image is binarized to get a binary image; 2) Use a horizontal projection profile of the binary image to estimate the width of the bounding box of Chinese character; 3) Combined with character recognizer, use a vertical projection profile to adaptively find vertical cutting positions; 4) Shortest-path searching method is adopted to find nonlinear cutting paths to remove noise components in the segments. Experimental results demonstrate the effectiveness of the proposed method in comparison with other related work in the literature.