医学图像检索中基于加权马氏距离的相关反馈方法

Medical Image Retrieval Relevance Feedback Method Based on Weighted Mahalanobis Distance

针对医学数字图像,提出了一种将动态权重调整与加权马氏距离相结合的相关反馈方法.修改了检索策略,提高检索的查全率和查准率.克服了欧氏距离在计算相似度时特征向量间属性相关的问题,减少冗余,提高了检索精度.实验结果表明,该方法能有效利用用户反馈信息,改善检索性能.

In order to solve the ＂ semantic gap＂ between the underlying physical characteristics and high--level semantic image retrieval for medical digital image, a method of combining a dynamic weight adjustment and the weighted Mahalanobis distance was proposed to feedback. The search strategy was modified to improve the retrieval recall ratio and precision ratio. The method overcame the problems of the Euclidean distance in the calculation of the similarity of feature vectors between the property--related issues, reduced redundancy and improved the retrieval accuracy. The experimental results showed that this method could effectively use user＇s feedback to improve retrieval performance.