Research On Rotation Invariant Matching For Round Image

The matching technology on image is the core content in picture processing technology and machine vision detecting. In the system of the online measuring of industrial products, the traditional matching method is not applicable to angle rotation between the two matching images, While in round images on-line detection system, there is always existence some angle deviation between the collecting images and benchmark, which brings some difficulties to the detection work. The issue is proposed based on this background, with the example of beer cap quality inspection, aiming to rotation changes of angle between two images this paper was studied.This articles has done some research about the collected round image, and has carried out depth study for the preprocessing and matching of round image. The main research contents as follows.In the pretreatment stage, the paper has made some studies about the location for the collected cap image. Making the round image isolated from the background which based on the mathematics theory of three points positing a circular by segmented linear transformation.Then proposed two angle correction algorithms for image rotation that based on key point detection and Fourier transform. Using the adaptive corner detector based on curvature scale space algorithm to detect corner points, to select a stable key point in the corners which have been detected. Connection the selected point with the geometric center of the image in order to certain the angle of the rotation image. The latter Fourier transform algorithm was introduced, training its time-frequency domain as invariance to rotation, firstly the image rotation angle was corrected by using Fourier transform, which made the untested image and reference image in the same direction.At last using the fast matching algorithm that based on sub block mean and the method of round image matching based on log-polar transform to match the untested and reference image. The experimental results show that compared with existing matching algorithms, the proposed algorithm can solve the rotation with any angle between two images' matching, and the algorithm has a high matching accuracy and applicability. In addition, the validity and reliability had got verified.