Abstract: In this paper, a new method of content-based image retrieval is proposed, which is based on improved color and texture features based on wavelet transform. Traditional image retrieval can not reflect the extent of importance of different objects of retrieval images. Firstly it extracts some regions of interest of the image, and assigns weights to different sizes by different degrees of interest. Then it extracts color and texture features, and calculates the similarity of two images by three different methods, including location similarity, ROI and retrieval image, retrieval example image and retrieval image, and takes the maximum similarity as the final similarity of the two images; it sorts the similarity of retrieval sample image and retrieval image of the database by size, and chooses the most similar image as the search results. Experimental results show that the method improves the image retrieval performance, reflecting the personalized retrieval, which has a good effect.

Key words: region of interest, feature histogram, wavelet transform, texture features, image similarity

摘要： 针对传统图像检索无法体现对检索示例图像中多个不同对象的检索要求程度的问题，提出一种改进颜色特征和小波变换纹理特征的图像检索方法。首先提取出图像的多个感兴趣区域，由感兴趣的不同程度分别赋予不同大小的权值；然后提取颜色特征和纹理特征，分别用对应位置相似度计算、感兴趣区域与检索数据库中图像整体的相似度计算和整体检索示例图像与检索图像数据库中图像相似度计算三种不同方法计算出两幅图像的相似度，取最大的相似度作为两幅图像的最终相似度；对检索示例图像与检索数据库中每个图像的相似度按大小进行排序，选择最相似的图像作为检索结果。实验结果表明，该方法提高了对图像检索的性能，体现了个性化检索，对图像检索具有很好的效果。

关键词: 感兴趣区域, 颜色直方图, 小波变换, 纹理特征, 图像相似度