A hyperspectral imaging system was used to extract image information from three seals with low recognition and serious picture coincidence in Shanghai Museum’s Qing Ming Shang He Tu. The problem of low recognition of one seal was solved by minimum noise fraction (MNF). For the other two seals, the problem of high coincidence with the picture was minimized by band clipping combined with MNF, which effectively separated the seals from the picture and achieved satisfactory effects. The results show that MNF can effectively improve the identification of the blurred seal and also effectively separate the seals from the picture when combined with the corresponding band clipping processing.