提取数字图像相对灰度特征的Robust估计及其快速算法

以玻璃饮料瓶清洁度的在线自动检验问题为背景，研究了应用计算机图像处理与 模式识别方法实现这一自动检验过程中的图像特征提取问题．文中提出了根据图像的 相对灰度检测方迹图像的思路，给出了描述相对灰度特征的参数和提取该参数的统计 估值方法．针对实际饮料瓶一致性差．图像复杂的特点，提出一种ｒｏｂｕｓｔ估计方法 来解决特征提取的可靠性问题．同时为了满足在线处理的实时性要求，文中在给出两 个有关数字图像相对灰度信号及其方差函数的定理的基础上，给出了一种实现ｒｏｂｕｓｔ 估计的快速算法，从而解决了特征提取中的可靠性与实时性之间的矛盾．文中方法已 在一套计算机图像处理系统上实现．结果完全满足实际要求．

Robust Estimation and Its Fast Algorithm for Extracting Relative Grey Level Feature of Digital Images

Taking thc automatic on-line inspection of cleanliness of glass beverage bottles as a background, this paper studies the image feature extraction which is one of problems to use computer image processing and pattern recognition methods for realizing the automatic inspection problem. The paper presents an idea of detecting blot images by using the relative grey levels (RGL) of an image, a parameter for expressing the RGL feature, and a statistical approach of estimating the parametern To accord with the irregularity in geometric shape of beverage bottles and the complexity of the bottle images, a robust estimation approach is proposed to obtain a reliable estimate of the RGL feature. To meet real-time requirement of the on-line processing, two theorems about the RGL and its variance, and a fast robust estimation algor-ithm based on the theorems are given in the paper, and thus the conflict between the reliability and real-time in the feature extraction procedure is solved. The approach presented in the paper has been implemented in a com-puter image processing system, and the experimental results are quite satis-factory with practical requirements.