Abstract: In order to more accurately describe image visual features, improve image retrieval efficiency, this paper proposes an image retrieval method based on hybrid features kernel. This method extracts image features of color, texture and SIFT, introduces Gaussian kernel function to establish image model of hybrid features kernel, and performs kernel K-means clustering in the high dimensional space. Experiments show that the hybrid model is better to describe image visual features than the single feature, improves the retrieval speed and accuracy.

Key words: image retrieval, hybrid features kernel, kernel-based clustering, Scale Invariance Feature Transform（SIFT） feature, color feature

摘要： 为了更准确地描述图像的视觉特征，提高图像检索的查准率与查全率，提出了一种基于混合特征核的图像检索方法。该方法提取图像的颜色、纹理、SIFT特征，引入高斯核函数，建立图像的混合特征核模型，在高维的核空间进行基于核的图像聚类。实验表明，该混合模型与传统多特征融合方法以及单一特征核方法相比，能够更好地表示图像的视觉特征，提高检索的查准率和查全率。

关键词: 图像检索, 混合特征核, 基于核的聚类, 尺度不变特征转换（SIFT）特征, 颜色特征