A method for filling holes in a naked-eye 3D multi-viewpoint image is provided, comprising: sequentially performing a step of mapping, determining and marking row by row, comprising mapping two first pixels adjacent to each other in a row direction among a plurality of first pixels in a reference camera image to two second pixels in a virtual camera image, determining whether there is a hole between the two second pixels, and if so, marking the hole; and sequentially performing a step of scanning, assessing and filling row by row, comprising scanning two second pixels adjacent to each other in a row direction, assessing whether there is a marked hole between the two adjacent second pixels, and filling the marked hole according to at least one non-zero second pixel which is adjacent to the marked hole.