基于多级降维处理的路面裂缝图像自动识别算法

Automatic Recognition Algorithm of Pavement Cracking Using Image Processing Based on Multi-layer Dimensionality Reduction

针对路面裂缝图像识别难以适应图像背景的复杂性、去噪效果不佳和存在较多断续边缘的情况,提出了基于多级降维处理的自动识别算法,主要有5个部分:(1)融合灰度和梯度信息的图像预处理;(2)局部OTSU分割;(3)基于连通性检测的二值图像去噪;(4)基于8方向Sobel梯度的像素连接;(5)8×8裂缝子块识别与去噪。随后,测试实例显示,算法通过5个分辨率层级的图像处理,可获得完整而连续的裂缝图像。最后,针对10张512×512的路面破损图片,对全局OTSU分割、8方向Sobel检测、Canny算子和文中算法进行测试,各算法综合性能指标F1值依次为77.31%、19.58%、20.21%和91.04%,准确率依次为68.10%、25.82%、38.57%和85.78%,召回率依次为89.74%、16.04%、14.25%和97.71%,体现了文中算法的优越性。

Aiming at the fact that image recognition of pavement cracking tends to have a failure in fitting for the complexity of image background, a poor denoising effect and broken edges, an automatic image recognition algorithm based on multi-layer dimensionality reduction is proposed, which has five main parts. （1） image pre-processing combining grayscale with gradient information; （2） local OTSU segmentation; （3） binary image denoising based on connectivity checking; （4） pixel connecting on the basis of 8-direction Sobel gradients; and （5） recognition and denoising of 8×8 cracking subimages. Then, test example shows that the algorithm can obtain unbroken and continuous cracks through processing of images with 5 resolution grades. At last, performance tests are carried out on ten 512 × 512 pavement images for global OTSU segmentation, 8-direction Sobel detection, Canny detection and the algorithm proposed above, from which the F1 （comprehensive performance index） values come out as 77.31%, 19.58%, 20.21% and 91.04%, Precisions are 68. 10%, 25.82%, 38.57% and 85.78%, and Recalls are 89.74%, 16.04%, 14.25% and 97.71%, for each of the algorithms sequentially, which indicates a outstanding performance of this algorithm.