In naked eye 3D display, the traditional multi-viewpoint acquisition method is a uniform acquisition. The study shows that the viewpoint distribution constructed by the display is not uniform due to the aberrations of the column lens grating, and there are contradictions between the distribution of the collected viewpoints and display, producing problems such as misaligned viewpoint images and incorrect perspective relationships, which affect the final viewing experience. In this paper, a multi-viewpoint image distribution correction scheme is proposed for the above problems. The scheme consists of a viewpoint screening algorithm and an intermediate viewpoints generation network. Combined with the real distribution law of spatial viewpoints, the viewpoint screening is performed for the uniformly collected multi-viewpoint images to guide the intermediate viewpoint prediction network to generate virtual viewpoints at corresponding positions to match the display construction viewpoint positions. Experimental results prove the scheme effectively solves the mismatch between the spatial viewpoint of the display and the position of the acquisition viewpoint, thus improves the viewing quality of the naked eye 3D display.