Abstract: The present invention provides a method for automatically cropping a scanned image having irregular boundaries. A pattern sheet is provided on the bottom surface side of the cover of the optical scanner. The pattern sheet is defined by multiple pattern units for serving as referential coordinates. When the scanner is initialized, the pattern sheet is pre-scanned to generate a pre-scanned image. The pre-scanned image of the pattern sheet will be analyzed to get the referential data, including the index, the referential intensity level, and the pixel-positions of each pattern unit. The referential data of each pattern unit will be stored in a database. When auto-cropping an area of interest (AOI) from a pre-scanned image, the pre-scanned image of the original will be logically divided into multiple document blocks according to the positions of the pattern units. Then, compare each document block with its correspondent pattern unit to determine if the document block contains AOI data.