关键词: 相关反馈, 检索结果集, 移动查询点算法, SVM方法

Abstract: Most of the relevance feedback techniques for image retrieval focus on adjusting user’s feedback or perfecting the retrieval method used by the system. Unfortunately, rich information contained in retrieval results per round is neglected. In this paper, based on query point movement algorithm, we define the retrieval set that can be got by truncating part of the retrieval results, then a novel algorithm using SVM method combining with memorization which was left in the system’s interactive progress to deal with the retrieval results was propose. The experimental results show that our method is more effective and more accurate than traditional ones.

Key words: relevance feedback, retrieval set, query point movement algorithm, SVM method