结合卷积与转置卷积特征的模糊车牌复原方法

深度学习算法在图像去噪领域已经得到了很好的效果;但目前对于深度学习算法在模糊图像复原领域的研究没有更深入的研究。直接应用图像去噪的方法对模糊车牌进行复原实际上可行的,但会产生复原图像细节缺失,时间代价高的缺点。针对这些问题,吸取去噪方法的优点,提出将原始图像信息与转置卷积复原后的图像信息相结合的方法,重新构建了图像复原网络结构;并根据图像特点自定了损失函数。实验通过与已有的方法进行对比说明,提出的复原方法在复原车牌图像质量上和复原效率上都有很好的表现;同时对模糊运动角度与不同噪声具有健壮性;而模糊运动像素越大的图片,复原图像的质量也会下降。

Blurred License Plate Recovery Method Based on Convolution Features and Transpose Convolution Features

Deep learning algorithm in the field of image denoising has been very good results, but for the depth learning algorithm in the field of blurred image restoration without further study. It is practically feasible to directly apply the image denoising method to recover the blurred License plate, but it will have the disadvantage of missing the detail of the restored image and high time cost. Aiming at these problems, taking the advantages of denoising method, this paper proposes a method of combining the original image information with the reconstructed image information, reconstructs the image restoration network structure, and customizes the loss function according to the image features. Compared with the existing methods, the experimental results show that the proposed restoration method performs well both in the quality of reconstructed images and in the recovery efficiency, and at the same time is robust to ambiguous motion angles and different noises. However, the larger the ambiguous motion pixels Picture, the quality of the restored image will also be reduced.