基于小波纹理、语义特征和相关反馈的医学图像检索

Medical image retrieval based on wavelet texture, semantic feature and relevance feedback

本文针对胸片和胸部CT扫描两种医学图像库,提出一种融合小波纹理、语义特征和相关反馈的医学图像检索技术。首先分析比较了纹理的三大描述方法:灰度层共现矩阵,纹理谱,小波变换,采用了频谱法中的一种基于小波变换的纹理描述方法进行纹理分析。为了进一步提高检索精度,将语义特征融合到该算法中,然后利用相关反馈技术有效地提高检索准确率。通过在原型系统上的实验,从查准率、查全率、平均序号验证了此方法的有效性。

We propose a medical image retrieval method based on the combination of wavelet texture, semantic feature and relevance feedback for two kinds of medical image databases, which are sternums and chest CT images. First, we analyze three methods of texture descriptor： wavelet transform, texture spectrum and gray level co-occurrence matrix, and utilize the method of medical image retrieval by wavelet transform. In order to improve the retrieval accuracies, semantic feature is combined with wavelet transform. Then the technique of relevance feedback is used in the algorithm to enhance the effectiveness of retrieval. Finally, a simple prototype system is developed to compare the precision, the recall rate and the average serial number by three experiments. Experimental results show that the proposed approach is effective.