摘要
提出了一种基于自组织映射神经网络的图像检索算法,通过有效地融合图像的颜色特征和纹理特征从图像库中查找与示例图像相似的图像。对于颜色特征,本算法将图像中各像素的R,G,B颜色作为输入值,对颜色相似的像素进行聚类,并将聚类结果映射成二维映射图。二维映射图中每个阶的像素数目作为特征1;每阶中像素的平均坐标作为特征2。为了增强对图像的描述能力,利用Jhanwar等人提出共现矩阵作为改进的纹理特征,该特征作为特征3。相比已有方法,本文算法获得了更好的图像检索性能。
This paper presents a self-organizing map neural network based image retrieval algorithm.For color features,the proposed algorithm utilize R,G,B color of each pixel as input values,and then cluster these pixels according to color similarity.Afterwards,clustering results are mapped into a two-dimensional map.The number of pixels in each order of the two-dimensional map is used as feature 1,and the average coordinate of pixels in each order is utilized as feature 2.To enhance image description ability,co-occurrence matrix is regarded as feature 3.Next,we linearly combine the above three features to calculate image similarity.Experimental results show that our algorithm can effectively retrieve images accurately.
出处
《科技通报》
北大核心
2013年第2期55-57,共3页
Bulletin of Science and Technology
基金
咸阳师范学院引进人才项目(3908XSYK339)
陕西省教育厅专项科研计划项目(09JK811)
咸阳师范学院校内基金项目(201002005)
关键词
图像检索
神经网络
自组织映射
颜色特征
纹理特征
image retrieval
neural network
self-organizing map
color feature
texture feature
