Research on Color Normalization of Traffic Sign Image Based on Surface Segmentation and Improved K-means Clustering Algorithm

Yang Liu,Yang Liu,Wei Zhong
DOI: https://doi.org/10.1145/3424978.3425155
2020-10-15
Abstract:In this work, a color normalization method is shown. Firstly, on the basis of analyzing the characteristics of traffic sign color combination, the classification of traffic sign image based on traffic sign color is demonstrated, and the color normalized mapping based on the matrix of traffic sign and color is analyzed, and the mapping target color set of traffic sign color normalization is resolved. Secondly, based on the construction of color classifier based on surface segmentation, the traffic sign color is initially classified, and then the classification of non-reserved colors is realized by using the improved K-means clustering algorithm. According to the analysis of K-means clustering algorithm, the improvement is proposed and analyzed. One is the self-selection of initial clustering centers based on color class screening by surface segmentation and the other one is the distance measurement calculation based on the clustering center axis, by which the similarity of the sample is measured by the distance from the sample to the clustering center axis. The experimental verification shows that the color normalization model proposed in this paper is simple and effective, and can well realize the color normalization of traffic signs.
What problem does this paper attempt to address?