Based on the different ability for detecting haze of each spectral band of remote sensing multispectral image, the blue spectral band which is more sensitive to haze is used to label the haze and non-haze images. The red spectral band which is less sensitive to the haze and the mean-shift segmentation are used to label bright object. Bright object can correct and compensate the haze images before dehazing processing. Taking domestic high-resolution remote sensing image as test object, the correction effect is evaluated based on the quality parameters of high-resolution remote sensing images. The research result shows that all image quality parameters of each spectral band are improved after correction, and image features in non-haze area keep consistency before and after correction. Detailed information of image in haze area is recovered and image quality is significantly improved after correction.