The invention discloses a natural image segmentation method based on label meaning inference, belongs to the field of digital image processing and relates to image segmentation tasks. The invention provides the natural image automatic segmentation method. The method is characterized in that segmentation areas of a digitally-marked image is utilized to automatically estimate the number of image areas after segmentation, each area is enabled to express an independent object, and the continuity among the areas is not broken, so that the understanding requirements of mankind perception are better met. On the premise that the high similarity in the areas and high association among the segmentation areas are ensured, a more meaningful image segmentation result is obtained.