Abstract: Image stitching can enlarge the range of viewing angles and increase different images information, and it is used in many fields such as industry, civil, and military. However, smog weather is an environmental problem in our country, because it can cause serious degradation of images. The loss of characteristic information will have negative impacts on the subsequent stitching process. Firstly, the smog image should be improved. In this paper, the application of the Multi-Scale Retinex (MSR) algorithm and the comparison and objective evaluation between it and the Histogram Equalization (HE) is discussed. Then, after removing the smog, the image is registered using local invariant features and the Speeded-up Robust Features (SURF) algorithm, and the Euclidean distance is adopted to obtain a satisfactory matching. Finally, the image stitching after registration may produce discontinuity of brightness in the overlapping area, and a higher quality stitching image can be achieved more quickly by using the progressive fade-out method. Through experiments and simulations, the smog images could be well stitched after removing the smog.

Key words: multi-scale retinex, image stitching, speeded-up robust features, image fusion