优化的邻近支持向量机在图像检索中的应用

Application of Optimized Proximal Support Vector Machine in Image Retrieval

邻近支持向量机由支持向量机衍生而来，它将支持向量机中二次规划问题的求解转换为线性方程组的求解，从而能在保证一定精度的情况下更加快速地得到分类器。传统的非线性核邻近支持向量机不能很好地解决多范围数据的多分类问题。提出了一种邻近支持向量机的优化方法，并将其应用到图像检索中。它利用高斯函数将图像特征数据映射到0-1之间以提高其差异化水平，并将其放入非线性核中，然后以加权K-means聚类算法选择最优参数，从而提高了非线性核PSVM的分类能力。实验以coral图像库中的4类图片作为图片库，对比了优化前后的检索命中率。实验结果表明：优化后的检索效果优于优化前，说明将优化的邻近支持向量机应用于图像检索是有效的。

Proximal support vector machine （PSVM） is derived from support vector machine （SVM） and converts the quadratic programming problem into linear equations, so that it can ensure the classifier more quickly under the condition of certain precision. The multi-class classification problem of multi range data can not be solved well by the original nonlinear kernel PSVM. This paper presents an optimization method for PSVM, and applied the optimized algorithm to image retrieval. The Gauss function is used to image feature data map to the range of 0 N 1 to enhance their difference level, then the different data is put into nonlinear kernel function, finally weighted K-means clustering algorithmis used to select the optimal parameters of PSVM. Experiments are carried out on 4 types of images from coral image database as the picture library, and the hit rate is compared between the original PS- VM and the optimized algorithm. Experiments show that the performance of optimized PSVM is better than original algorithm, and it is effective to use the optimized PSVM into image retrieval.