Abstract: When the visual SLAM (simultaneous localization and mapping) method is applied to underwater environment, the interference caused by raised sediments makes it difficult to extract and track SLAM feature points, and the uneven illumination by artificial light sources causes the uneven distribution and small number of feature points. To solve those problems, a semi-mean dust-filtering and illumination equalization based feature enhancement algorithm is designed for underwater images. According to the pixel characteristics of the impurities in water, the semi-mean filter algorithm removes the raised sediment in the image from outside to inside in the order of detection-filtering. And, the distribution of pixels in areas with sufficient and even illumination is counted, and a law is obtained that the environmental characteristics at different locations in the same terrain is similar. Based on the law, the underwater illumination model is solved to restore the raw image into an image with even illumination, and thus image features are enhanced to extract more effective feature points. Various underwater terrain datasets are processed by the filtering and enhancement algorithm, and some tests are carried out with the ORB-SLAM3 algorithm. The results show that the number of feature points extracted and the number of point clouds for mapping are increased by 200% in average by using the filtered and enhanced datasets. So, the image dust-filtering and feature enhancement algorithm can effectively improve the performance and stability of visual SLAM algorithms.