基于霍特林变换的稻米彩色图像粒形检测算法  

A Hotelling Transform-Based Algorithm for Rice Grain Type Detection with Color Images

作　　者：刘璎瑛[1] 丁为民[1] 陈建伟[2] 沈明霞[1] 

机构地区：[1]南京农业大学工学院,江苏南京210031 [2]江苏省粮食局粮油质量监测所,江苏南京210011

出　　处：《中国水稻科学》2010年第3期325-328,共4页Chinese Journal of Rice Science

基　　金：国家863计划资助项目(2008AA100903);江苏省科技支撑计划资助项目(BE2008401)

摘　　要：利用稻米分割后轮廓灰度图与背景灰度图的灰度均值之差和灰度方差之差进行米粒图像分割效果定量评价,对7个彩色通道的稻米图像进行分割评判,选取I1(红色、绿色、蓝色通道的平均值)通道进行稻米图像分割。提取分割后标注的单粒米粒边界的二维坐标向量,对坐标向量进行霍特林变换,通过计算变换后米粒最小外接矩阵来表征稻米粒形,简化了现有的稻米粒形检测算法。检测稻米粒型时,算法在MATLAB7.5.0环境下运行。该算法所得米粒长宽比与人工检测结果的平均相对误差为1.65%,每幅图像平均耗时0.323s;而最小外接矩形算法的平均相对误差为2.24%,每幅图像平均耗时2·837s。An algorithm for rice grain type detection using color image segmentation and the Hotelling transform was proposed. The rice grain images were segmented in I_1 band（the average of red, green and blue color bands） chosen from seven color bands according to quantitative analysis of segmentation performance. The coordinate vector of segmented image edge was converted by the Hotelling transform. Grain size features were extracted in the new coordinate system using minimum enclosing rectangle. This algorithm was run in MATLAB 7.5.0 to count rice grain ratio. The relative error compared with manual measurement were 1.65% and 2.24%, and the running times were 0.323 s and 2.837 s per image, respectively, for this algorithm and algorithm using minimum enclosing rectangle.

关 键 词：霍特林变换 稻米 彩色图像 颜色通道 粒形检测 

分 类 号：S126[农业科学—农业基础科学] S511.033