机器学习的光照不均图像边缘检测系统

针对传统图像检测系统检测质量较低,容易出现图像边缘细节模糊的现象,提出基于机器学习的光照不均图像边缘检测系统。构建图像边缘检测系统总体框架,检测系统主要分成图像收集、低级处理、高级处理以及检测结果的后续处理四部分,以该系统框架为基础采用支持向量机模式将正常样本空间内不能被线性分类或者近似性分类的图像,通过非线性映射至高维特征空间,利用回归训练的核函数替代内积运算,即可完成非线性拟合。根据拟合结果将图像内的每个像素点,都作为一个训练样本,使二维图像可以用对应的拉格朗日函数表示,同时也可以表示成核函数线性组合,最后通过零交叉检测法即可完成图像边缘检测。通过实验证明,所提系统能够很好地完成光照不均图像边缘检测,且目标细节清晰。

The edge detection system of uneven light image based on machine learning

In view of the low detection quality of traditional image detection system,which is prone to fuzzy edge details,a machine learning-based edge detection system for uneven illumination image is proposed.The overall framework of image edge detection system is constructed.The detection system is mainly divided into four parts:image collection,low-level processing,high-level processing and subsequent processing of detection results.Based on the framework,the support vector machine model is used to map the images that cannot be classified linearly or approximately in the standard sample space to the high-dimensional feature space through nonlinear mapping and regression training.The nonlinear fitting can be completed by replacing inner product operation with a kernel function.According to the fitting results,every pixel in the image is regarded as a training sample,so that the two-dimensional image can be expressed by the corresponding Lagrangian function,and can also be expressed as a linear combination of kernel functions.Finally,the image edge detection can be completed by zero-cross detection.The experimental results show that the proposed system can detect the edge of the uneven illumination image well,and the details of the target are precise.