The invention discloses a light field image mosaic method based on super pixels, which comprises the following steps: inputting light field data to be mosaic to obtain light field sub-aperture images,extracting feature points, matching and screening, and predicting a global homography transformation matrix according to the feature points; the subaperture image of light field being segmented intosuperpixels, and the homography transformation matrix corresponding to each superpixel being predicted according to the feature points under the current depth hypothesis, and the light field being mapped according to the homography transformation matrix, and the optimal depth hypothesis being evaluated and selected as the superpixel hierarchy map; the weight matrix of local mapping being established, and then the optimal homography transformation matrix of each super pixel being predicted according to the global homography transformation matrix and the weight matrix of local mapping, and the light field being mapped according to the optimal homography transformation matrix of each super pixel; the light field being fused to get the result of light field splicing. The invention solves the problems of misalignment and ghost caused by large parallax variation, and realizes the accurate parallax tolerance light field splicing method.