The invention discloses a novel light field image depth flow estimation method. The method comprises: step one, original light field data are processed into one group of sub-aperture image arrays; step two, a spatial derivative of a center sub-aperture image is calculated; step three, bilateral filtering is carried out on each sub-aperture image; step four, an initial light field depth flow parallax estimation value is estimated; step five, iterative optimization is carried out on the initial parallax estimation value obtained at the step four; and step six, the parallax estimation value is normalized, wherein the parallax range is set to be in an interval [0,1]; the parallaxes are fused with a voting mechanism, statistics of the obtained parallaxes is carried out by using a histogram andthe parallax with the largest frequency number is used as a final parallax estimation value. According to the invention, on the basis of the Horn and Schunck optical flow algorithms, a depth parameterbeta is introduced into an HS optical flow calculation formula directly and creatively, so that the depth estimation accuracy is improved.