一种新型的基于神经网络的图像检索算法

A Novel Method of Neural-Network-Based Image Retrieval Using Multiple Features

分析传统基于多特征的图像检索算法,针对于传统算法核心仅放在如何更多的提取图像的底层特征,却忽视了如何组织这些特征,提出一种改进的基于神经网络的图像检索算法。即在深入提取图像颜色以及形状特征的同时,将图像检索中特征权值调整问题转化为最优化问题,进一步利用神经网络求取权重的最优解,使检索效果达到最优。仿真实验表明,该算法能够准确和高效地查找出所需的图像,具有很强的鲁棒性和检索能力。

According to the analysis of traditional arithmetic of multi-character image retrieval, we propose a novel arithmetic of image retrieval based on neural network because of the traditional arithmetic attach importance to the feature extraction but not understand the importance of how to organize these feature. We use two feature, HIS color information and CSS（Curvature Scale Space） as shape information. And then use neural network to adjust the weigh of the feature information. As a result, we show that good results can be obtained by two features and neural network. When we use that arithmetic, we could find the image that we need exactly and efficiently.