In this paper, an offline palmprint image is divided into four components: blank regions, knuckle-finger regions, high quality palm regions and unrecoverable low quality palm regions. The high quality palm regions are the ROI of offline palmprint images and the other three regions constitute background regions. A block-based segmentation algorithm is proposed to identify the high quality palm regions from background regions. In experiment section, the segmentation algorithm is tested from three aspects: by human inspection, by comparing the segmentation results with an official pattern and by observing the change of accuracy rate of minutiae after masking the unrecoverable low quality palm regions. Our palmprint database for testing contains 200 image samples and each image comprises 1600 blocks. Only 2.4% of blocks are misclassified. The accuracy rate also increases significantly after masking.