一种基于神经网络提取图像特征的图像检索方法

Image retrieval based on feature extraction by neural network

脉冲耦合神经网络是新一代的人工神经网络,具有优良的自适应图像分割和自适应图像特征提取能力。本文有机的运用自适应图像分割和自适应图像特征提取,提出了一种新的图像特征———空间自适应类直方图,它既包含了图像的空间位置特征,又包含了图像的灰度特征。最后,我们把空间自适应类直方图与传统的基于共生矩阵的图像纹理特征组合作为图像的特征应用于图像检索。大量实验表明本文方法的有效性。

Pulse coupled neural network （PCNN） is a new generation of artificial neural networks and is characterized with adaptive image separation and adaptive image feature extraction. This paper presents a novel image feature, Space Adaptive Histogram（SAH）, which is obtained the output of PCNN stimulated by the image, to depict a histogram- like features of an image but with both spatial and intensity features of the image. Then PCNN is utilized for adaptive image separation, and for SAH feature extraction based on the separated parts of the image. Finally, the SAH feature is combined with the conventional texture feature of comatrix to realize image retrieval. A large number of experiments on image retrieval together with the comparison with the conventional comatrix method shows the effectiveness of the method presented here.