【摘要】 图片分类可用作图像搜索引擎的预滤波,以降低图像检索时的图像匹配数量,提高检索速度。本文提出了一种基于分形特征———局部分维数变化率(LFDS)的图片分类算法,该算法不需要任何的图像先验知识,仅利用分形特征就可将自然景物的照片和人工绘制/计算机生成的图形区分出来。系统随机测试了445幅大小从192×128到2012×3094不等的图片,该算法对图形库的分类准确率为91·71%,图像库分类准确率为85·25%,实验结果证明了该算法的快速和有效性。更多还原

【Abstract】 The image classification can be used as a pre-filter of the image search engine to reduce the amount of image matching, and speed up the image index processing. Without any prior knowledge of image content, a fast and effective method based on a fractal feature-Local Fractal Dimension Slope(LFDS ) is presented to classify photographs taken from natural scenery and graphics drawn by hand or generated by computer. The system tested on 445 random images whose size ranged from 192×128 to 2 012×3 094. The graphics can be classified correctly at the rate of 91.71% and the classification accuracy of photograph is 85.25%. The experimental results have demonstrated the effectiveness and high-speed performance of the proposed method in this paper.更多还原