自定义模糊逻辑与GAN在图像高光处理中的研究

Research on Custom Fuzzy Logic and Generative Adversarial Networks in Image Highlight Processing

本文提出一种新的关于图像高光修复的方法.使用模糊逻辑对图像高光区域进行划分定位,然后使用添加了扩张卷积的卷积神经网络对图像进行修复.在该方法中,为了保证图像补全区域与原图像的一致性,我们使用了全局判别器网络和局部判别器网络相结合的方式来区分真实图像和补全图像.全局鉴别器判断整个图像,以评估修复后图像从整体上观测是否是连贯的,而局部判别器只则判断在高光区域部分所生成的补丁图像,该模块确保生成图像在局部区域的一致性.两者相互结合以达到优化模型的作用,最后为了使生成图像部分与原图像的融合更加自然,再通过图像融合技术使生成部分与原图片进一步融合,使得修复效果更进一步.

This paper put forward a new method for repairing the highlight of image.Fuzzy logic is used to divide and locate the highlight of the image,and then use dilated convolution to repair the image.In this method,in order to ensure the consistency of the image completion area and the original image,we use a combination of global discriminator network and local discriminator network to distinguish the real image from the complete image.The global discriminator examines the whole image to assess whether the restored image is coherent as a whole,while the local discriminator only judges the patch image generated in the highlight area of image.This module ensures the consistency of the generated image in the local area.The two are combined with each other to achieve the role of optimizing the model.Finally,in order to make the fusion of the generated image part and the original image more natural,the fusion part is further merged with the original image through image fusion technology,so that the repair effect is further.