The paper aims to research and improve the printing overprint deviation detection algorithm based on cross line alignment identifier to improve the accuracy of the deviation detection. Firstly, the K-means clustering algorithm was used to cluster the C, M, Y and K images obtained after color separation. Secondly, the fourimage arerespective segmented and the each foreground pixels were extracted. Then the foreground cross lines in C, M and Y images were respectively subtracted from the cross lines in K images to eliminate the interference of four-color overlapping regions. Finally, the center of the cross line was extracted and the deviation was calculated. Experiments on 15 printed image samples showed that results of the improved overprint deviation detection algorithm based K-means clustering cross line was closer to the manual measurement results than Otsu method, and the error was within (±0.04 mm). The cross-line segmentation method based on K-means clustering is more resistant to overlay interference than Otsu method in foreground cross-line extraction.