The invention belongs to the technical field of 3D display, and particularly discloses a visual comfort prediction and non-linear mapping visual comfort enhancement method based on saliency self-adaptability. The method includes the implementation steps that firstly, a color
saliency map and a
parallax saliency map of images are extracted; secondly, the color
saliency map is normalized, a normalization value serves as a weight and is multiplied by the
parallax saliency map, and then a saliency self-adaptability
parallax map is acquired; thirdly, saliency discomfort regions are extracted through threshold segmentation; fourthly, parallax characteristics are extracted, visual comfort is estimated based on SVR, VC values are calculated, and
visualization operation is conducted; fifthly, non-linear mapping is conducted on the saliency self-adaptability parallax map; sixthly, through the DIBR technology, the mapped parallax map and an original right view serve as input, and virtual images are synthesized. The predicted visual comfort is higher in relevance compared with visual comfort acquired through subjective evaluation, excessive parallax is reduced, saliency information of the images is kept, the visual comfort of the three-dimensional images is improved, and the method can be widely used for 3D display.