Near-infrared subcutaneous vein segmentation method based on multi-feature clustering

The invention designs a near-infrared subcutaneous vein segmentation method based on multi-feature clustering, and can realize the multi-feature extraction and the automatic clustering of a vein. The method comprises the following steps:1) adopting a NiBlack and a morphology algorithm to realize the segmentation of a skin area and edge mirror extension; 2) obtaining a vein similarity diagram, a vein directional diagram, a vein scale diagram and an initial segmentation vein through multiscale IUWT (Isotropic Undecimated Wavelet Transform) and Hessian matrix analysis; 3) extracting and repairing a vein branch center line by adopting the initial segmentation vein and the vein directional diagram, and correcting the position and the direction of the branch center line by adopting a segmentation spline fitting method; 4) on the basis of a vein branch direction, calculating a coordinate mapping relationship between an artwork and a branch outline image, and extracting normalized second-order Gaussian characteristics and vein similarity characteristics after an IUWT enhanced image and the vein similarity image are independently mapped to outline image space; and 5) utilizing the obtained vein characteristics to cluster the outline image into three types including skin, vein and a fuzzy region by adopting a K-means algorithm.