A false matching removal method based on superpixel motion statistics

Abstract

The invention discloses a false matching removal method based on superpixel motion statistics. The method comprises the following steps: carrying out feature extraction, description and matching on two images to be matched; Segmenting the to-be-matched images I1 and I2 by using an improved superpixel segmentation algorithm to obtain two superpixel mark graphs; And establishing a superpixel motionstatistical model based on the superpixel marker map, and realizing automatic screening of registration feature points of the non-rigid deformation image through the model. According to the method, asuper-pixel segmentation strategy is adopted to replace simple rectangular grid division, super-pixel blocks obtained through segmentation are tightly connected in space, colors and textures in singlesuper-pixel blocks are kept consistent, a segmentation result better follows the moving edge of an object, and therefore it is guaranteed that feature points in super-pixels have the same or consistent moving tendency.

Description

technical field [0001] The invention relates to the field of computer image matching, in particular to a method for removing false matching based on superpixel motion statistics, which can be used to remove false feature matching of non-rigid deformed images. Background technique [0002] The fundamental purpose of the feature point registration method is to establish the matching correspondence between two or more image feature point sets. Due to the influence of lighting conditions, noise, geometric transformation, space warping, etc., it is very challenging to achieve completely accurate feature matching. At present, the feature operators used for image registration mainly include SIFT operator (scale invariant feature transformation), SURF operator (accelerated robust feature), and ORB operator [1] Wait. Using the above method can achieve relatively accurate feature registration, but when non-rigid deformation or large-scale displacement occurs between images, it is ea...

AI Technical Summary

Problems solved by technology

[0004] However, studies have shown that although the matrix grid is simple, it often contains different texture components. Therefore, it is often not accurate enough to assume the same motion law in the same rectangular grid, and certain statistical errors will also occur.

Images