柑橘表面缺陷图像快速准确分割方法

摘要

Citrus surface defects affects the quality of fruit and food safety, so the detection of citrus surface defects has a great significance for improving the quality and value of fruits. Local Binary Fitting (LBF) is a image segmentation model which based on Chan-Vese (CV) model. Because the traditional LBF model has high requirements on the initial contour line and poor anti-noise ability. This paper presents a new LBF model based on the original LBF model by adding a kernel function (Gaussian function) and linear level set method for the LBF model improving. In order to solve the problem of image segmentation on the common defects of citrus surface (insect pests, decay, anthrax, wounds), an improved LBF model was used to verify whether the improved LBF model effectively extract the four common defects of cit rus surface. The results showed that the improved LBF model could be quickly identify the surface defects of insect pests, decayed fruits, anthrax fruits and medicinal fruits. The result is great and can be obtained with the defect image level set evolutionary images as well. It has the advantages of fewer iterations, shorter segmentation time, insensitive to the initial contour position, more smooth and complete segmentation contour, and accurate recognition of defect boundaries, which effectively solves the problem of the traditional LBF model. The experimental results showed that the improved LBF model was suitable for the segmentation and extraction of four kinds of citrus surface defects, which is feasible, rapid and accurate, and also provide a reference for the identification of citrus surface defects and on-line detection of citrus.%柑橘表面缺陷会严重影响水果的品质和食用安全,柑橘表面缺陷进行检测对于提高水果品质、提升水果价值有着重要意义.LBF(local binary fitting)是一种基于Chan-Vese(CV)模型的局部化的图像分割模型.由于传统的LBF模型存在对于初始轮廓线的位置要求高且抗噪能力差,对于灰度不均匀图像分割效果欠佳的问题,通过在原LBF模型基础上,添加一个核函数(高斯函数)和线性水平集的方法,对LBF模型进行了改进.针对柑橘表面常见缺陷(虫伤、腐烂、炭疽、药伤)的图像分割问题,采用改进的LBF模型进行试验,来验证改进后的LBF模型对柑橘表面四种常见缺陷能否进行有效的分割提取.通过对虫伤果、腐烂果、炭疽果、药伤果四组样本分别进行分组试验,结果表明:改进后的LBF模型对于虫伤果、腐烂果、炭疽果、药伤果的表面缺陷能够进行快速的识别,分割效果好并能得到与缺陷图像相对应的水平集演化图像.具有迭代次数少、分割时间短、对初始轮廓位置不敏感、分割轮廓线更加光滑和完整、缺陷边界识别准确等优点,有效地解决了传统LBF模型的不足.试验验证了改进后的LBF模型适用于四种柑橘表面缺陷的分割提取,具有可行性、快速性和准确性,为柑橘表面缺陷的识别与柑橘在线检测提供参考.