融合高阶多尺度特征的路面裂缝图像分割算法

Pavement Crack Image Segmentation Algorithm Based on High-order Multi-scale Features

针对裂缝图像的光照不均匀、斑马线等复杂背景使得传统的裂缝图像分割算法容易丢失细微及末梢裂缝等问题,本文提出了一种基于泰勒级数和多尺度特征的裂缝图像分割算法.首先,采用瑞利和高斯分布构成的有限混合模型对裂缝背景和目标进行建模,并使用期望最大化算法求解混合模型的参数;然后,通过泰勒级数的展开式描述裂缝的梯度方向,利用尺度变换构造裂缝图像的高阶多尺度特征;最后,将灰度有限混合模型和裂缝高阶多尺度特征融合到马尔科夫随机场模型,通过条件迭代算法优化求解裂缝标号场最大后验概率来实现图像分割.性能测试和不同算法对比分析实验表明,本文算法在保证裂缝几何参数不变的前提下能够抑制非裂缝目标并保留低对比度、细微和末梢裂缝,分割准确率达到85.93%、灵敏度达63.87%,衡量指标优于其他算法.

Aiming at the problems of uneven illumination,zebra crossings and other complex backgrounds of crack images,which make the traditional crack image segmentation algorithms easy to lose subtle and terminal cracks,this paper proposed a crack image segmentation algorithm based on Taylor series and multi-scale features.Firstly,a finite mixture model composed of Rayleigh and Gaussian distributions were used to model the crack background and target,and the expectation maximization algorithm was adopted to solve the parameters of the mixture model;Then,the gradient direction of the crack was described through the expansion of Taylor series,and scale transformation was applied to construct the high-order multi-scale features of the crack image;Finally,the gray-scale finite mixed model and the crack high-order multi-scale features were fused into the Markov random field model,and the maximum posterior probability of the crack label field was optimized by the conditional iterative algorithm to achieve image segmentation.Performance testing and comparison and analysis experiments of different algorithms show that the algorithm in this paper can suppress non-fracture targets and retain low contrast,fine and distal cracks under the premise of keeping the geometric parameters of the fractures unchanged.The segmentation accuracy rate reaches 85.93%,and the sensitivity reaches 63.87%.The indicator is better than other algorithms.