Method of standard feature descriptor technology based on robustness overlapping

Abstract

The invention discloses a method of a standard feature descriptor technology based on robustness overlapping, and relates to the technical field of standard feature descriptors. According to the method, firstly, a second-order standard partial derivative is calculated based on a Scharr operator so as to enhance the invariance of the descriptor and avoid the influence of Gaussian filtering on the operation efficiency; then an overlapping description grid is constructed to improve the continuity of the description grid and enhance the discrimination of the descriptor; and finally, neighborhood response weighted summation and normalization are carried out to generate a 64-dimensional descriptor with illumination invariance. Verification is carried out by measuring a recall ratio-debugging rate curve and a time-consuming experiment, and compared with a traditional floating point type descriptor, the robustness of the robustness overlapped standard feature descriptor is higher.

Description

technical field [0001] The present invention relates to the technical field of feature descriptors, in particular to a method based on robust overlapping standard feature descriptor techniques. Background technique [0002] Feature descriptors are algorithms that determine the corresponding positions of features in different images. It mainly solves the problem of distinction and invariance between features under viewpoint transformation, blur transformation, illumination transformation, JPEG compression transformation, scale transformation and rotation transformation, which is the prerequisite for feature matching and feature tracking. Descriptors maintain the distinguishability and invariance of features, overcome the impact of rotation, scale and affine deformation on recognition, and thus improve the matching or tracking effect. Therefore, it is of great significance to study descriptors. [0003] In recent years, descriptors are mainly divided into two types: binary an...

AI Technical Summary

Problems solved by technology

The binary descriptor has a great advantage in the efficiency of generation and matching, but the discrimination of the descriptor generated simply by comparing the gray value of the pixel is poor, especially when the image has a viewing angle change, the performance of the descriptor will be seriously degraded

However, the accuracy of the descriptor based on wavelet response calculation is poor, and it does not solve the problem of grid discontinuity, so there is still room for improvement in the robustness of the descriptor

[0005] Due to the above-mentioned traditional descriptors, there are still the following problems in the feature description: using the wavelet response to approximate the gradient to describe the feature is not robust; The descriptor block strategy destroys the continuity of the grid and reduces the robustness; the more dimensions of the feature vector will greatly reduce the matching operation efficiency

Images