摘要
提出了一种新的图像特征表示方法,首先提取图像的底层颜色信息获取颜色特征值,通过对图像中物体的边缘检测计算像素点的边缘方向角度值,并对颜色特征值和边缘方向角度值进行量化。然后根据相邻像素点之间量化结果的数值分析,为每个像素点建立8维特征向量。再以中心像素点与相邻像素点间不同的位置关系为基础,为每种位置关系赋予不同的权重,根据像素点的特征向量计算出图像中每一个像素点的特征值。最后统计图像中具有相同特征值的像素点个数,形成特征直方图,以此作为图像检索的依据。实验表明本文方法能够有效描述图像的颜色分布和图像中物体的空间结构,更加细致地记录图像信息,进一步增强图像之间的区分能力。与其他方法相比,本文方法检索效果更好。
Abstract: A novel method is presented for image feature representation. The underlying color information of images is extracted to obtain color feature values based on the presented method. With the edge detec- tion of objects in images, edge orientation values of pixels in images are calculated. Then color feature values and edge orientation values are quantized. Based on quantization result analysis of neighboring pix- els, an eight dimensional feature vector is constructed for each pixel. Different weights are given to dif- ferent positions based on the relationship between a central pixel and neighboring pixels. According to feature vectors of pixels, characteristic values are calculated. Finally, the pixels number with the same characteristic value is counted to form a histogram, which is the basis for image retrieval. Experimental results demonstrate that the proposed method can effectively describe color distribution and spatial struc- ture of objects in images. Moreover, more image details can be saved to enhance the discrimination pow- er. The method is proved to be much more effective than other methods.
出处
《数据采集与处理》
CSCD
北大核心
2016年第3期577-583,共7页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(61170145
61373081
61401260)资助项目
教育部博士点基金(20113704110001)资助项目
山东省自然科学基金(ZR2010FM021)资助项目
山东省科技攻关计划(2013GGX10125)资助项目