Random sampling consensus algorithm for image recognition based on voting decision and least squares

Abstract

A random sampling consensus algorithm for image recognition based on voting decision and least squares method. The voting decision method is introduced to find the most likely correct point pair, and the rotation and translation transformation matrix is ​​calculated by the principle of least squares method. The rotation and translation transformation matrix calculated by the invention is not only accurate, fast, but also very stable, and the calculation result will not fluctuate randomly each time; the invention not only optimizes the image comparison method, but also helps the subsequent learning of the data processing chip The feature information of the reference image is continuously increased, thereby effectively improving the false rejection rate and making the data processing chip more intelligent; the present invention also realizes image stitching more efficiently and accurately through the continuously updated rotation and translation transformation matrix.

Description

technical field [0001] The invention relates to a data processing method, in particular to a data processing method for image comparison and image stitching. Background technique [0002] In order to realize image comparison, and perform image splicing according to the image comparison result, the prior art uses the basic image of image comparison as a template image, and uses the image for comparison as the image to be tested, the template image and the feature points of the image to be tested. After Hamming matching and spatial de-pseudo of the descriptor of , a number of rough matching point pairs are obtained, and the Random Sampling Consensus algorithm, which is abbreviated as RANSAC in English, is used to remove mismatches and calculate the rotation and translation transformation matrix. The traditional random sampling consensus RANSAC idea is to estimate the H matrix by arbitrarily selecting three pairs of points in the set of coarse matching point pairs, and then use...

AI Technical Summary

Problems solved by technology

[0003] Because the traditional random sampling consistent RANSAC algorithm randomly selects point pairs each time to estimate the H matrix, the randomness is relatively strong, the amount of calculation is large, and the efficiency is low. It needs many iterations to find a suitable H matrix, especially when the input The larger the number of error point pairs in the two matching point sets of the random sampling consistent RANSAC algorithm and the larger the proportion, the greater the number of iterations required, otherwise the H matrix may not be found, but the greater the number of iterations, the greater the number of iterations. The computing time and storage capacity will increase proportionally, and the H matrix and consistent set calculated after each random sampling consistent RANSAC algorithm are not exactly the same

Images