Image Dehazing Method Based on Multi-scale Feature Fusion

Abstract

In recent years, image dehazing has become an active topic in computer vision.This paper presents a novel dehazing method based on adaptive clustering and image fusion for restoring a single image degraded by fog or haze.Firstly, the coarse-scale transmission map is addressed based on adaptive clustering for the haze image.Then, the saturation of haze image is extracted as fine-scale transmission map, which can reflect the scene depth information truly and naturally.Next, the accurate transmission map is generated by using Laplacian pyramid fusion technique.Finally, the haze-free image is restored by solving the image degradation physical model.The effectiveness of the proposed method is evaluated on various real-scene haze images.Experimental results indicate that our proposed method is simpler and more effective compared with several well-known methods, and even achieve better visual results.Specifically, our method is more fast and suitable for real-time requirement.