Recognition of pavement damage types based on features fusion

摘要

In order to automatically identify pavement damage types, firstly, four features extraction methods, such as Contourlet Transform, Edge Direction Histogram, Gradient Direction Histogram, and Stratified Gradient Direction Histogram, were analyzed in this paper. Secondly, a new features extraction method based on combination features of Contourlet Transform and Edge Direction Histogram are supposed. Thirdly, based on the image database of Southeast University, the SVMs classifier using linear kernel function was used to conduct the single feature experiments, and the corresponding recognition rates were 84.32%, 75.14%, 51.93%, 73.25%. Finally, the experiments of combined features of Contourlet Transform and Edge Orientation Histogram were conducted and the experimental results show that the combination of the two features can achieve a more detailed description of the image features, thereby increasing the image recognition classification rate.