提出了直接从H.264 I帧编码码流中提取块边缘模式特征的算法，并检验了该特征在图像检索中的应用。该算法利用H.264 码流中得到的帧内预测模式和预测系数提取出五种边缘模式特征，该特征符合MPEG-7中所定义的边缘描述符，适合基于内容的图像检索算法。实验结果表明，采用该特征进行图像检索具有较高的检索准确率，相对于只采用Intra预测模式纹理描述符的方法，平均查准率提高了5.07％。

This paper proposes a feature extraction algorithm for the block edge pattern from H. 264 I-frame bit streams and verifies the extracted features by image retrieval applications. In the proposed algorithm, five block edge patterns are extracted based on the intra prediction mode and the prediction coefficients from H. 264 bit streams, which accord with the edge descriptor in MPEG-7 and are suitable for content-based image retrieval algorithms. Experimental results show that the image retrieval based on the extracted features is relatively precise, and the average precision is 5.07% higher than that of the retrieval based only on the texture descriptor of intra prediction mode.