The invention discloses a
light field image multi-dimensional
feature extraction and modeling method for an
image enhancement task. The method comprises the following steps: S1, inputting sub-view image features of a
light field image; s2, the sub-view image features are input into a space
branch, a
polar plane branch and an angle
branch in parallel, the space branch is used for extracting and enhancing space details and textures in all view angles, the
polar plane branch is used for conducting directional modeling on a polar
line structure and
parallax information and achieving weighted integration through depth fusion, and the angle branch is used for conducting directional modeling on the polar
line structure and
parallax information; the angle branch is used for carrying out consistency modeling between visual angles; and S3, fusing the features output by the spatial branch, the
polar plane branch and the angle branch to obtain a
light field output feature. According to the method, the spatial detail
recovery and the angle consistency maintenance of the light field image can be efficiently realized, the polar
line structure is prevented from being damaged, the calculation efficiency is optimized, and the real-
time processing requirement is met.