Abstract: An algorithm is proposed in the paper to segment caption region.By the algorithm,firstly, the caption features are extracted by wavelet transformation and reconstruction on the image,and secondly, statistic features are calculated block by block.At last blocks are classified by K-mean clustering method.In comparison with othe r algorithms,the algorithm is simpler and no requirement for setting any threshold. Experimental results show that the proposed algorithm performs well,even for captions with unknown font, size,and position.