改进的综合BTC颜色矩和DT-CWT的图像检索算法

Improved image retrieval algorithm combining BTC color moment and DT-CWT

特征提取是基于内容的图像检索中的关键技术。针对基于单一特征检索效果不理想的问题,提出一种改进的综合颜色和纹理特征的图像检索算法。该算法在YIQ颜色空间中进行特征提取,首先结合方块编码(BTC)的思想,提取颜色矩作为颜色特征;采用双树复小波变换(DT-CWT)提取纹理特征,融合两种特征并利用相似性度量方式进行图像检索。实验结果表明算法所提取的颜色、纹理特征更利于检索,使用综合特征检索的平均查准率比同类算法更高。

The feature extraction is the key technology of content-based image retrieval. In view of the problem that the image retrieval has not an ideal effect when basing on single feature, an improved image retrieval algorithm combining color feature and texture feature is proposed. In the algoritbm,the image is firstly converted into YIQ color space to extract features. The color moments are calculated as the color feature applying the thought of block truncation coding（BTC） , the dual-tree complex wavelet transform is proposed to extract texture feature in each YIQ color component. The extracted features are synthesized and the method of similarity measurement is taken advantage of to retrieve images. The conclusions are drawn from experimental results that the color and texture feature extracted in the algorithm is beneficial to image retrieval, and the algorithm using comprehensive features has better retrieval performance compared with other algorithms.