Abstract: Concerning the low recognition rate of vehicle-logo images which were shot outdoor under moving conditions and with a variety of qualities, a new recognition method was proposed based on image quality and Principal Component Analysis (PCA) subspace. Firstly, the image fuzzy degree was computed according to fuzzy theory. Then the training samples were divided into different subsets in terms of fuzzy degree and their corresponding PCA subspaces were constructed. At last, the vehicle-logo was recognized in the specific subspaces which were chosen by the vehicle-logos fuzzy degree. The experimental data show that the reconstruction errors based on this method are less than those of traditional PCA subspace, thus the proposed method has more powerful capability in pattern depiction and can obtain higher recognition rate. The comparative test results further indicate the validity of this method.

Key words: Intelligent Transportation System(ITS), vehicle-logo recognition, image quality, Principal Component Analysis (PCA)

摘要： 针对室外动态获取的车标图像质量差异大而导致的识别率不高的情况，提出了一种结合图像质量的主成分分析子空间的车标识别方法。该方法首先基于模糊理论计算车标图像的模糊度，进而根据模糊度将训练样本分成不同的子集并生成相应的PCA子空间族，最后根据待识别车标图像的模糊度选择相应的子空间族进行识别。实验数据表明基于模糊度PCA子空间进行的重构误差比基于传统PCA子空间进行的重构误差小，因此其模式描述能力强，从而获得较高的识别率。与其他算法的对比实验进一步表明该算法的有效性。

关键词: 智能交通系统, 车标识别, 图像质量, 主成分分析