基于小波域的深度增强车牌图像去雾算法设计

Design of Defogging Aalgorith for Depth-Enhanced License Plate Image Based on Wavelet Domain

为解决雾天拍摄的车牌图像边缘模糊、色彩失真的问题,提出了端到端的基于小波域的深度增强车牌雾图去雾算法。先根据大气散射模型构建了雾天车牌图像数据集,之后利用小波变换将车牌雾图从空间域转换成小波域分量图像,最后将处理后的小波域分量图像进行逆小波变换,重构出的干净车牌图像。去雾网络以U-Net的编解码结构为主体框架,通过多个残差组从训练集中提取特征,并在解码器中引入“SOS”深度增强策略对编码器和下层输入的特征进行融合和细化,用以提高去雾车牌图像的峰值信噪比。实验表明,上述网络在结构相似度和峰值信噪比上具有明显优势,在处理合成车牌雾图和实际拍摄的车牌雾图上,去雾效果表现良好。

In order to solve the problem of blurred edges and color distortion of license plate images taken in foggy days, an end-to-end depth-enhanced license plate fog image defogging algorithm based on wavelet domain is proposed. First, we built a foggy license plate image data set based on the atmospheric scattering model, then used wavelet transform to convert the license plate fog image from spatial domain to wavelet domain component image, and finally we performed inverse wavelet transform on the processed wavelet domain component image to reconstruct the image Clean license plate image. The defogging network used the U-Net codec structure as the main framework, extracted features from the training set through multiple residual groups, and introduced the SOS enhancement strategy in the decoder to fuse and refine the features of the encoder and the lower layer input. It was used to improve the peak signal-to-noise ratio of the defogging license plate image. The experiments show that the network has obvious advantages in structural similarity and peak signal-to-noise ratio. It has a good dehazing effect in processing synthetic license plate fog images and actual license plate fog images.