SIFT (Scale Invariant Feature Transform) feature point gradient generation method based on calculation manner optimization

Abstract

The invention discloses an SIFT (Scale Invariant Feature Transform) feature point gradient generation method based on calculation manner optimization. The method comprises the steps of: dividing feature point neighborhood gradient directions according to a circle and angle subdivision; calculating gradient values of Lx and Ly of x and y directions; calculating the gradient values of Lx and Ly anda gradient ratio values needed for direction judgment; and calculating a gradient quadrant, judging a gradient direction, and assigning values to the gradient direction and gradient intensity. According to the SIFT feature point gradient generation method based on calculation manner optimization provided by the invention, a shape of SIFT feature point neighborhood division in the image is changed,thus an original square neighborhood is replaced by a circular field, thus a calculation manner of gradient generation is changed, ingenious fixed-point operations are used to avoid complex floating-point operations, higher-precision approximation results are calculated and derived, thus resource consumption and calculation time consumption of hardware realization are greatly reduced, and at thesame time, rotation invariance of an image matching system is greatly improved when an image rotation angle is small. The method has broad application prospects.

Description

Technical field [0001] The invention relates to an image feature point gradient generation method based on a SIFT algorithm family, and in particular to a SIFT feature point gradient generation method based on optimization of a calculation method. Background technique [0002] The feature descriptor is used to characterize the feature point information and its uniqueness, usually a multi-dimensional vector. The SIFT feature descriptor is a descriptor based on the local information of the image, that is, a descriptor expressing the uniqueness of the feature point is generated from the image information of the feature point neighborhood, and the SIFT feature descriptor has translation, rotation and scale invariance, and A certain degree of illumination and affine robustness and many other excellent properties [1] . [0003] Rotation invariance is one of the important properties of the SIFT descriptor. The SIFT algorithm determines its main direction by counting the peaks of the grad...

AI Technical Summary

Problems solved by technology

[0009] (1) From the calculation formulas of the above gradient value and gradient direction, it can be seen that it includes complex floating-point operations such as division, square root, and arctan (tangent), especially the square root operation not only consumes a lot of resources, but also requires a lot of time delay cycle

Therefore, these operations are not suitable for implementation in hardware such as FPGA due to resource consumption and long calculation time.

[0010] (2) Due to the rough merge method of the gradient direction, in the actual image processing, the matching effect is better only when the image is rotated by a multiple of 45°, and the image is rotated by a small angle (such as 11.25°, 22.5°) or non- When the multiple of 45°, the rotation invariance of the image is poor, which affects the matching performance of the image system

Images