Disclosed are a method and a device for discriminating a boundary of image, and a display panel, for effectively discriminating whether an image has a boundary and in which direction the boundary is. The method for discriminating the boundary of image comprises: receiving an image information to be discriminated to form a matrix of grayscale parameter values, and dividing, with a grayscale parameter value corresponding to a sub-image unit to be processed as a center, the matrix of grayscale parameter values into a n×n matrix and a (n+2)×(n+2) matrix (S101); determining respectively in the n×n matrix and the (n+2)×(n+2) matrix: a minimum gradient and a minimum standard deviation in a row direction, in a column direction, in a first diagonal direction, and in a second diagonal direction, dispersion with respect to the minimum standard deviation and dispersion with respect to the minimum gradient (S102; S103); outputting a first code value when the determined dispersion is greater than N multiples of a minimum corresponding thereto, outputting a second code value when the determined dispersion is smaller than the N multiples of the minimum corresponding thereto (S104); determining, based on an outputted code value, whether the image to be discriminated has a boundary and in which direction the boundary is (S105).