(2) Gray Stretch

The image of the iced insulator has the characteristics of fuzziness and noise in the image background. The direct equalization method was selected to make a gray value of the image uniform distribution for enhancing contrast and highlighting the iced insulator image details. The result is shown in Figure 3.

**Figure 3.** The grayscale image and the image after direct equalization and its histogram of grayscale distribution. (**a**) Original grayscale image and its histogram. (**b**) Grayscale image and its histogram after direct equalization.

(3) Image denoise

In fact, after the direct equalization processing, the noise interference is still present in the iced insulator image. To remove the small bright spot and improve the definition of image, the median filtering algorithm method was chosen to diminish the gap of the image. The iced insulator image edge can be sharpened, and the obvious background noise can be decreased through enhancing the filtering effect. The result is shown in Figure 4.

**Figure 4.** The image with noise interference and the image after median filtering processing. (**a**) The image with noise interference. (**b**) The image after median filtering processing.


The key of image processing is image segmentation, which is to segmen<sup>t</sup> an image into meaningful regions by extracting some target area of image characteristics, and then obtain the binarization image [16,17]. The maximum entropy threshold segmentation algorithm was proposed for acquiring excellent efficacy of segmentation and the characteristics of recorded images during the ice regime. This method is essentially using the images' regional features to segmen<sup>t</sup> images based on the similarity between the pixels. The higher entropy value of the segmentation image can indicate the more information it contains, and it is beneficial to the e ffect of division [18]. The calculation formula of entropy is

$$H(S) = -P\_1 \ln P\_1 - P\_0 \ln P\_0 \tag{2}$$

where *H*(*s*) is the statistical value of the amount of information that the binarization image contained after segmentation. *P*1 and *P*0 represent the probability that the output value of the segmentation image is one and zero, respectively.

According to the e ffective segment, the gray value of the image is compressed and transformed into 0 or 255 pixel values. Then, the approximate edge contours are obtained and the process of binarization is completed. The image processing result is shown in Figure 5. It can be observed that the maximum entropy threshold segmentation method can extract the object points of the image and remove the redundant information. Therefore, the proposed algorithm ensures the segmentation more efficient and segments insulator images with intensity inhomogeneity correctly.

**Figure 5.** The binarization image after maximum entropy threshold segmentation processing.
