基于深度注意力机制的低照度图像增强方法

Low-light Image'Enhancement Method Based on Deep A ttention Mechanism

复杂环境下的低照度图像具有光照分布不均、多光源叠加作用等特点,导致增强后的图像真实性不足、图像噪声增加等问题。针对低照度图像的特点,提出了一种基于深度注意力机制的低照度图像增强方法。设计生成对抗全局自注意力低照度增强网络(GSLE-GAN)以实现低照度图像的增强。在生成器中设计并使用注意力模块,提高模型对于光照分布特点的提取能力以及生成图像的真实性,采用局部鉴别器与全局鉴别器共同作用的方式使图像具有更丰富的细节信息,使用非配对数据及对模型进行训练,以提升模型的鲁棒性并进一步保证生成图像的真实性。通过对比实验,证明了文中所提方法的优越性,并在目标检测任务中证明了方法的有效性。

Low-light image in complex environment has the characteristics of uneven light distribution and multiple light sources superposition,which leads to the problems of insufficient authenticity of the enhanced image and the increasement of image noise.According to the characteristics of low-light image,a low-ight image enhancement method based on deep attention mechanism is proposed.An adversary network GSLE-GAN is designed and generated to enhance the low-light image.In the generator,the attention module is designed and used to improve the model's ability to extract the characteristics of the light distribution and the authenticity of the generated image.The local discriminator and the global discriminator is used to work together to make the image have more detailed information,non-paired data is used and the model is trained to improve the robustness of the model and further ensure the authenticity of the generated image.The advantages of the method proposed are proved through comparative experiments,and the effectiveness of the method is proved in the target detection task.