The surface defect image of paper plastic composite bag surface defect is affected by noise, uneven illumination and the characteristics of the defect itself. It will lead to over segmentation or under segmentation in image defect region segmentation. In view of this phenomenon, a segmentation algorithm of paper plastic composite bag surface defect image which combined edge detection and adaptive region growth method was proposed. Firstly, Sobel operator and morphological operation were used to segment the defect image after bilateral filtering for the first time; then, the minimum circumscribed rectangle was marked on the defect area and its shape feature was calculated, and whether to continue segmentation was determined by determining the size of the shape feature; finally, the centroid of the defect region in the image that conformed to the continuous segmentation was used as the initial seed point to carry out self adaptation on the original image. The second segmentation result was formed by growing the region to complete the defect image segmentation. Compared with other algorithms, this algorithm can achieve better segmentation results for all kinds of common defects, and the Dice coefficients was above 0.93. The algorithm has high segmentation accuracy and strong robustness, which can meet the needs of industrial production.