Abstract: The tread image segmentation is the process of realizing the separation of tread area and
background, which is the link between image preprocessing and tread image defect detection. Aimed
at the problem of missing image information, low precision of regional contour segmentation and
poor anti-interference ability, a method of color tread image segmentation based on improving
watershed algorithm is presented in this paper. Firstly, the proportion of the incident component of the
tread image with the reflective component and the RGB three color channel is adjusted by using the
multi-scale retinal enhancement (MSRCR) with color restoration. Then the color image gradient
graph is calculated directly. The color gradient image foreground and background mark is completed
by improving RGB color component fusion operation and then the initial segmentation result is
obtained by the watershed transformation. Finally, combined with the tread contour azimuth
characteristic, image connected domain extraction segmentation algorithm is designed and the tread
surface extraction is completed. The experimental results show that the method is good to divide the
edges of the image, the color fidelity, the fog haze, the illumination disturbance ability, and obtain the
ideal wheel tread segmentation results.

Key words: tread segmentation, color image, MSRCR algorithm, watershed algorithm, marker extraction