中文摘要：

针对现有搜索引擎算法不能完整依据用户的查询需求,导致检索质量差的问题,提出一种基于聚类和用户兴趣模型的个性化车辆图像搜索算法,实现个性化搜索。根据用户感兴趣目标特点,选择目标多特征集合;采用多核线性融合方法计算相似度;在此基础上,提出使用优化降维支撑向量机进行基于多核的动态聚类,建立用户兴趣模型,并将个性化的搜索结果返回给用户。研究结果表明：与传统的搜索算法相比,新算法增加了用户的参与方式,解决了底层视觉特征与高层语义间的鸿沟问题,能够明显提高平均查全率和查准率;新算法比基于颜色的传统搜索算法的平均查全率和查准率分别提升了9%和24.6%,比基于纹理的传统搜索算法的平均查全率和查准率分别提升了28%和42.6%。

英文摘要：

Traditional search engines can＇t completely satisfy users~ searching needs which will lead to declined refrieval quality and increased costs. The paper proposed a personalized vehicle image searching algorithm based on clustering analysis and user interest model. According to the characteristics of users~ interested objects, Multi-feature sets were first selected for object repre- sentation. Then multi-kernel linear confusion was adopted to calculate the similarity. Then an optimal SVM was utilized to complement the multi-kernel clustering, build a specific user interest model and return the personalized searching results to the users. The analysis of experiment re- sults indicates that compared with the traditional searching algorithm the improved algorithm can enhance participation methods of users and solve the gap problem between low-level vision feature and high-level semantics. It can increase the mean recall and precision ratio respectively by 9% and 24% compared with the traditional one based on color, and by 28% and 42. 6% compared with the traditional method based on terfure. 1 tab, 4 figs, 8 refs.