In the seal verification system of bank, the precondition of identifying the seal on notes correctly is the seal Abstracted must be with high quality and low noise. On account of those facts such as the complexity of note background, the shade of inkpad, the randomicity when people seal and the noise cause by sampling, we have more difficulties in stamp Abstracting and mending. After analyzing both the advantages and disadvantages of the RGB and HIS color space model, the authors build a new stamp Abstraction model and propose a new fill algorithm. Experiments indicate that with this Abstraction model they can Abstract the seal with high enough quality and reduce the noise to the least at the same time, and with this fill algorithm they can restore seal better and avoid over-fill at the joint of stroke. So, the requirements of the subsequent seal imprint verification can be met.