The paper aims to propose an image restoration scheme based on improved K-means clustering to solve the problem of poor restoration effect caused by poor separation effect of the target to be repaired in the common subregional image restoration algorithms. Firstly, the image to be segmented was transformed into CIELab color space, and K clustering centers were obtained by clustering a and b components. The rough segmentation results were obtained by changing the number of clustering iterations. Then, the segmentation results were refined by mathematical morphology, and the target object and background were separated accurately. Finally, the Reinhard algorithm was used to migrate the color of the target and background respectively, and the image restoration results were obtained. The region segmentation algorithm in the model proposed in the article had better separation effects than the classic watershed algorithm, the maximum between-class variance method and the largest between-class variance algorithm based on the Lab channel. The image restoration result using the Reinhard color migration algorithm was closer to the ideal restoration effect. From the final results, it can be concluded that the overall effect of the proposed restoration method is better than that of the traditional subregional image restoration algorithm. It can provide the necessary theoretical basis for production practice.