Image Dust Storm Synthetic Method Based on Optical Model

To address the problem that the dusty image dataset is small and difficult to collect, this paper presents a synthetic method for generating dusty image based on a classical optical model. The proposed method first learns the physical process of generating dusty image according to the classical optical model. Then, the transmission map is estimated and combined with the presupposed dust storm color map as inputs for obtaining a synthetic dusty image. Finally, considering the impact of image scene depth on the synthesis of dusty image, the proposed method selects an appropriate value of input parameter to obtain final synthetic dusty image. Experimental results on an image dataset captured in clear weather show that the synthetic dusty images obtained by the proposed method can be used as a good substitute for real dusty images.