基于Zernike矩和BP网络的道路交通标志识别方法研究

道路交通标志的背景相当复杂，颜色失真严重并存在不同程度的几何失真现象。不变矩是图像的一种统计特征，具有平移不变性、旋转不变性和比例缩放不变性，被广泛的应用于图像识别中。在研究了Hu矩和Zerni—ke矩基础上，提出基于Zernike矩与BP网络相结合的道路交通标志识别方法。识别过程分别对图像进行了Hu矩和Zernike矩特征提取、BP网络训练与测试、对形变图像进行分类识别。结果表明：基于Zernike矩和BP网络的交通标志识别方法具有很强的抗图像平移、缩放和旋转识别能力，实现简单、训练速度快、识别率高等特点，且识别准确率优于Hu不变矩目标自动识别。

Research on Method of Traffic Signs Recognition Based on Hu Invariant Moments and Zernike Invariant Moment

During real time recognition process of the traffic signs, there are color and geometric dis- tortions because of complicated background noise. Invariant moment is a statistical property of images and widely used in the image recognition which possesses image＇s translation, scaling and rotation invariance. Based on the research of Hu and Zernike invariant moments, algorithm of traffic signs recognition based on Zernike invariant moments and BP neural network is presented in the paper. First, the invariable moment eigenvector of the traffic signs images is extracted as eigenvalue. Second, BP network is trained and test- ed. At last, the deformed traffic signs images are recognized by using the trained BP network. The experi- mental results show that this method is based on Zernike moments and BP neural network possesses image＇s translation, scaling and rotation invariance. And the method is simple, fast training and high rec- ognition rate in image＇s translation, scaling and rotation; and is prior to the automatic recognition method based on Hu invariant moments.