Quality inspection of package is vulnerable of illumination. Aiming at this problem,a defects detection algorithm based on the improved gradient magnitude(GM) similarity was proposed. Based on the existing defects detection method by gradient magnitude similarity, the local binary pattern (LBP) operator was integrated into the proposed algorithm.The algorithm utilized the characteristics of rotation invariance and gray scale invariance of LBP operator, and combines them with the gradient magnitude features of the image to be used in the defects detection of packaging, which improves the algorithm robustness to illumination.Experimental results show that the proposed algorithm has better anti-illumination ability in comparison with traditional methods,and under different illumination conditions, the detection accuracy of the proposed can reach 96.57%. Therefore, the proposed algorithm can be widely used in quality inspection of package and improve the defects detection accuracy.