The invention relates to an automobile surface image dust removing method based on dark channel prior. The automobile surface image dust removing method comprises the steps of obtaining an automobilesurface image to be dedusted; Judging whether an automobile in the image is white or not, if yes, reversing the image, and if not, keeping unchanged; Calculating a corresponding dark channel graph according to the established degradation model; Selecting the maximum value of the dark channel value as an ambient light value; Carrying out bilateral filtering processing on the dark channel image to obtain initial transmissivity, and refining the initial transmissivity by using guided filtering; Determining a dust-removed image according to the ambient light value and the transmissivity, determining the dust-removed image as a final dust-removed image if the dust-removed image is not a white automobile, and reversely rotating the dust-removed image if the dust-removed image is the final dust-removed image; And performing quality evaluation on the dust-removed image. On the basis of dark channel prior, image defogging is achieved, the visual effect is good, high image definition and contrast are kept, and later application of a machine vision algorithm for automobile surface scratch image recognition and color feature extraction is facilitated.