基于差分卷积的弱光照车牌图像增强

Weak License Plate Image Enhancement Via Differential Convolution

车牌识别作为现代化智能交通系统中重要的环节,对提升路网效率以及缓解城市交通压力等问题具有重要的社会意义,然而弱光照车牌图像识别仍然具有重大的挑战。构建了一个基于差分卷积神经网络的弱光照车牌图像增强网络,将车牌的纹理信息解耦为水平垂直和对角线两个方向,对不同尺度空间的低照度图像进行纹理增强。为了避免增强结果局部过曝或低曝,该方法使用YCbCr颜色空间的损失函数来优化模型。图像增强实验结果表明,所提出的方法较传统的低照度图像增强方法相比,图像客观质量结果峰值信噪比提升了0.47 dB。同时,在仿真车牌和真实场景的车牌识别实验结果也证明了所提算法对于低照度图像感知质量提升的有效性。

Objectives:As an essential part of the modern intelligent transportation system,license plate recognition is of great social significance to improve the efficiency of the road networks and alleviate urban traffic pressure.however,it is still a great challenge to weak illumination license plate image recognition algorithm.Methods:A weak illumination license plate image enhancement network based on differential convolution neural network is constructed,the texture information of the license plate is decoupled into horizontal vertical and diagonal directions,and the texture of low illumination images in different scale-spaces is enhanced.In order to avoid local overexposure or low exposure of the enhancement results,this method uses the loss function of the YCbCr color space to optimize the model.Results:The results of image enhancement experiments show that compared with the traditional low-intensity image enhancement methods,the objective image quality result peak signal-to-noise ratio improves 0.47 dB.Conclusions:At the same time,the experimental results also prove the effectiveness of this algorithm in recognition of composite license plates and real scene license plates for the improvement of low-illumination image perception quality.