The debonding defect is a crucial indicator affecting the physical safety performance of packing materials, and computed tomography (CT) stands as an effective method for its detection. However, due to the debonding defects’ close to the outer contour and small area, low contrast, they are susceptible to interfere from other information during segmentation, making traditional active contour models less suitable. In order to segment the packing materials debonding defects, SoftMax and regular term ChanVese (SPCV) model based on Chan-Vese is proposed by introducing SoftMax and regular term in this paper. Differences in segmentation effects between the SRCV model and various active contour models were systematically compared. Segmentation results on different images are utilized to demonstrate anti-interference capabilities and balance global and local information capabilities of SRCV model. The depth of debonding was measured using Euclidean distance. When the SRCV model is used to segment the debonding defects of packing materials, the segmentation curves are closer to the debonding edges and smoother. The accuracy and the Dice coefficient were 99.56% and 99.82%, respectively. And the error of the debonding depth was not more than 6%. The results show that the SRCV model is particularly suitable for the segmentation of tiny debonding defects with a large amount of interfering information, and has obvious advantages over other active contour models.