In order to improve the accuracy and reliability of defect detection for packaging cans production lines, an inner surface inspection system with a single camera was studied. By using morphological region extraction algorithms the inspection region of interest image can be obtained from the whole inner image. For defect feature detection an approach based on multivariate image analysis (MIA) was proposed. To cancel the effect of seam regions in the inner images, a method of images fusion was implemented to form the ring-like good sample image without seam region. By stacking both the ring-like good sample images and the test image, the multivariate test images were constructed with their overlapping part. By using MIA technique with principal component analysis (PCA), the principal component scores of the multivariate test images were obtained. As the feature space for defect detection the Q statistic image was derived from the residuals which were left after the extraction of the first PC and noise. The surface defects can be effectively detected using an appropriate threshold. The experimental results show that the proposed inspection system has less sensitivity to the inhomogeneous of illumination, and has more robustness and reliability with pseudo reject rate reducing to 2%.