结合稳定兴趣点和Gabor小波的图像检索

Image retrieval using stable interest points and Gabor wavelet

提出了一种基于稳定兴趣点和纹理特征的图像检索算法.该算法首先利用优化的Hessian检测器检测图像中的稳定兴趣点,并计算稳定兴趣点的环形邻域的伪泽尼克矩;然后,利用Gabor小波变换提取图像的纹理特征;最后,用不同图像伪泽尼克矩和纹理之间的差异来衡量图像的相似度,实现图像检索.实验结果表明,与其他基于兴趣点或者纹理特征的检索方法相比,该算法能够降低不稳定兴趣点的影响,有效提高了图像检索的准确率和查全率.

A new image retrieval method based on stable interest points and texture feature is proposed. Firstly,the optimal Hessian derivative filter is used to detect the stable interest points in the image.After that,pseudo-Zernike moments defined on the neighborhood of stable interest points in the annular region are calculated. Then, the texture feature is extracted by the Gabor wavelet transform. Finally, the difference in pseudo-Zernike moment and texture feature among images is used to depict image similarity. Experimental results show that this method reduces the effects of the unstable points and improves the image retrieval accuracy effectively comparied with other retrieval methods based on interest points.