不均匀光照和重影的仪表图像二值化方法

Binarization method of instrument image with uneven illumination and ghosting

  • 摘要: 高压计量仪表识别过程中需要对图像进行二值化操作,然而拍摄的仪表图像多出现光照不均和数字重影现象,导致传统方法对仪表图像的二值化困难。为此,提出一种基于卷积神经网络的二值化方法,用于对复杂光照下含数字重影的仪表图像二值化。该网络使用的数据集为真实环境下的仪表图像,首先对输入的图像进行降维提取特征,然后反卷积重建图像前景,最后输出二值图。将设计的网络与传统的二值化方法进行对比,实验结果表明,经该网络训练得到的二值图数字清晰且无重影,且测得的交并比(IoU)平均值为95.12,与样本标签图像的相似度最高,能够有效解决复杂环境下有重影的仪表图像二值化问题。

     

    Abstract: The image needs to be binarized during the identification process of the high-voltage meter. However, the phenomenon of uneven illumination and digital ghosting often appears in the instrument images, so that it is difficult to binarize the instrument images with traditional method. Therefore, a binarization method based on convolutional neural network was proposed to binarize the instrument images with digital ghosting under complex illumination. The data sets used in the network were the instrument images in real environment. Firstly, the dimensionality reduction was used to extract features of the input images, and then the foreground of images was reconstructed by deconvolution. Finally, the binary images were output by the network. Comparing the designed network with the traditional binarization method, the experimental results show that the binary images of the proposed network are clear and have no ghosting. The average IoU is 95.12, which is most similar to the sample label images. Therefore, the method can effectively solve the problem of binarization of instrument images with ghosting under complex environment.

     

/

返回文章
返回