Corner detection method based on neural network, storage medium and image processing system

Abstract

The invention provides a corner detection method based on a neural network, which comprises the following steps: S1, preprocessing the collected image; S2, extracting the edge of the preprocessed image; S3, cutting the image including the edge into a plurality of sub-images with pixel size of m * n according to a fixed step size, and recording the coordinate label of each sub-image in the originalimage; 4, carrying out dimensionality reduction processing on each sub-image by using dictionary learning to obtain a sparse representation of each sub-image; S5, respectively inputting a sparse representation of each sub-image to a pre-trained neural network, and judging whether there are corner points in each sub-image according to the output value of the neural network. The invention also provides a storage medium and an image processing system corresponding to the method. The invention still has high corner detection accuracy for noise interference, line segment interference and pixel high contrast.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a neural network-based corner detection method, a storage medium and an image processing system. Background technique [0002] The corner points in the image are stable and invariant under the affine transformation, and it can reflect a lot of useful information in the image, so it plays an extremely important role in feature recognition. Corner detection technology has a wide range of applications in camera calibration, image matching, motion estimation, panorama stitching, target detection and recognition, etc. The existing corner detection methods can be divided into two categories: the corner detection method based on image gray level and the corner detection method based on contour. [0003] Corner detection methods based on image grayscale currently mainly include Harris method, SUSAN method, etc. The Harris method is relatively simple, and the window W can be randomly sel...

AI Technical Summary

Problems solved by technology

However, the CSS corner detection method has several disadvantages: first, using the second derivative to calculate the curvature will be highly sensitive to local changes in the contour and noise; second, it is difficult to choose a suitable Gaussian scale to smooth the contour; third, how to choose the appropriate corner threshold

Later generations have improved the above defects, but there are still defects of too many adjustment parameters, and the problem of line segment interference cannot be solved

[0005] To sum up, the corner detection in the prior art has disadvantages such as poor anti-interference ability, low detection accuracy, too many adjustment parameters, and weak self-adaptation, which in turn causes problems in camera calibration, image matching, target detection and recognition. Errors and other issues

Images