基于全局和角点特征的图像检索

Image retrieval based on global and corner features

图像的颜色、纹理和形状等视觉特征是图像信息描述的重要内容,而这些特征是从图像的全局提取还是从局部提取,对图像的可区分性描述是不同的。为了更全面地描述图像信息以提高图像检索精度,从整幅图像中提取HSV直方图特征和LBP特征,然后提取图像角点的Hu矩形状特征和基于GLCM的纹理特征,融合这两类特征,选用相对曼哈顿距离进行相似性度量完成图像检索。实验结果表明,该图像检索方法的查准率有了一定的提高。

Color, texture and shape are important visual features in the description of image information. Either these features are extracted from the global or local image makes the different distinguishing description of the image. In order to describe image information more comprehensively and improve the accuracy of image retrieval, the HSV histogram feature and LBP feature are extracted from the whole image, then the Hu invariant moment feature and the texture feature based on GLCM are extracted from the corner of the image. Finally, the two kinds of features are combined to complete image retrieval by relative Manhattan distance to measure similarity. Experimental results show that this feature extraction method has better image retrieval accuracy.