A predetermined region set on a grayscale image is divided into rectangles each having a predetermined size (S706), and piled patterns which hold the accumulated values of pixel values at identical pixel positions upon piling up the divided rectangles is generated (S707). An alignment pattern is generated by calculating the logical sum of respective bit expression patterns (S708). A peak position which matches the alignment pattern on the piled patterns best is obtained (S710). The position of the predetermined region is corrected based on the peak position, and the predetermined region after correction is divided into rectangles each having the predetermined size (S711) Patterns in the divided rectangles are recognized (S712). Embedding information corresponding to each recognized pattern is specified (S713).