摘要
为解决在实际生产环境中使用计算机自动识别奶牛的问题．提出了针对奶牛图像的分割算法，以提高识别率。该算法首先将奶牛图像从RGB空间转化为Luv空间，然后将图像作4×4大小的子块划分，对各子块进行块内颜色聚类并提取颜色和纹理特征，通过颜色内聚将子块中的颜色范围缩小到2—4种，并将其中所占数量最多的两种颜色作为子块的颜色特征；纹理特征则是基于对子块分别横向和纵向扫描并统计出来的．反映子块的颜色在横向和纵向的分布特征。在此基础上按奶牛的颜色和纹理特征进行基于子块的区域生长，最终达到清晰分割的目的。在对50个实地采集样本中按本文所述的方法进行分割．分割准确率达到80％以上。实验证明，本文提出算法能有效地将奶牛从背景中分离出来。
This paper presents an algorithm for the segmentation of cow image to deal with the cow identification in the dairy factory. The cow image is firstly transformed from RGB space into Luv space, then the image is partitioned into sub-images which size is 4~4, and the color clustering has been done on each sub-image and the features of color and texture have been extracted. Colors in each sub-images have been reduced. Two main colors will be used to describe the main color feature of each sub-images. The texture feature which describes the distribution of the colors in each sub-images are counted by scanning the sub-images horizontally and vertically. The region growing algorithm has been done based on sub-images according to the cow features of color and texture. 50 pictures had been processed by the algorithm which presented by the paper, the accuracy rate of segmentation is 80% or higher. The experimental results show that the proposed algorithm can separate the cow from the background efficiently.
出处
《电子设计工程》
2011年第2期81-84,88,共5页
Electronic Design Engineering
基金
上海师范大学产学研项目(DCL200802)
关键词
彩色图像分割
颜色聚类
纹理特征
区域生长
多阈值
八连通域标记
color image segmentation
color clustering
texture feature
region growing
multi-threshold
8-connectedcomponents labeling
