The invention discloses a face portrait compositing method based on Bayesian inference, which mainly solves a problem that an existing method does not consider similarity constraints between adjacent image blocks in a neighbor finding or weight solving stage. The scheme of the face portrait composting method comprises the steps of 1, dividing a training portrait sample set, a training photo sample set and a test sample set; 2, dividing all images into image blocks and forming block sets; 3, dividing the training photo block set and the corresponding portrait block set into a plurality of subsets; 4, selecting nearest neighbor blocks from a training photo-portrait block set; 5, carrying out neighbor block reselection on the nearest neighbor blocks according to an Euclidean distance in the training photo-portrait block set, and solving a linear combination weight of the reselected neighbor blocks; 6, solving portrait blocks to be composited according to the neighbor blocks and the combination weight; and 7, iteratively executing the steps 5-6 for N times, and fusing to acquire a final composite portrait. The face portrait composting method has the advantages of high resolution of a composite result and more complete details, and can be applied to face retrieval and recognition.