Proposed is a human face resolution re-establishing method based on machine learning, which retains overall structure information about a human face in the process of realizing image resolution improvement, and avoids the occurrence of a local distortion in a generated output image. The human face resolution re-establishing method includes: acquiring an input image, the input image having a first resolution; based on the input image and a standard gradient image library having a second resolution, determining image gradient information about the input image; fusing the image gradient information, and superposing the gradient information obtained through fusion onto the input image; and generating an output image, the output image having a third resolution, wherein the second resolution and the third resolution are both greater than the first resolution.