A Conditional Generative Adversarial Network Method for Image Segmentation in Blurry Image

Nowadays, the wide application of machine vision is developing rapidly, and image segmentation is indeed one of the indispensable technologies. When the input image is blurred, the outline of the target object will be unclear, resulting in inaccurate segmentation. This paper proposes an image segmentation method based on Generative Adversarial Network, combined with Concatentation and Summation to enhance the encoder and decoder and improve these problems without preprocessing data. This paper uses two experimental methods to verify the final image segmentation effect. It uses the Intersection over Union (IoU) score as the evaluation metric. The number of low IoU in this model does not increase drastically with the degree of blur. Moreover, without an overfitting issue in the testing dataset. It can prove the universality of the model.