Image dehazing algorithm based on improved generative adversarial network

Abstract

Under foggy conditions, the images and videos collected by the device are blurred and poorly imaged, which greatly impacts the accuracy of subsequent visual tasks such as target detection and recognition. At present, the dehazing methods such as dark channel and AOD- Net based on estimating intermediate variables still have problems such as incomplete dehazing and large color error. Therefore, an image dehazing method based on the generative adversarial network is proposed. The generator adopts a dense block structure connected layer by layer to improve the details of the dehazed image. The discriminator uses PatchGAN to perform block determination and optimize the image resolution. Meanwhile, the generated dehazed image and the real fog-free image are trained and compared with the comparison method. The peak signal-to-noise ratio and structural similarity of the proposed method are improved on synthetic datasets, and the generated images retain better detail and clarity as observed by the human eye.