Quick calculation method based on Zernike moment of GPU

Abstract

The invention relates to a quick calculation method based on a Zernike moment of a GPU and belongs to the technical field of orthogonal moment acceleration. An eight-trigram limit symmetry algorithm of the storage radial direction multinomial coefficient by combining with the image repeated layout is provided, and the calculation speed of a random single moment can be remarkably improved by the mixing algorithm. A group packing scheme by combining the Block in a core Kernel is provided, and calculation of a facility or a set of moments is accelerated. The group packing scheme by combining cores can further shorten calculation time of moments of a set or a family of images, and can particularly overcome the defect that the mixing algorithm can be effectively conducted due to small calculation amount in small-sized images. According to the quick calculation method based on the Zernike moment of the GPU, development in the fields such as real-time image processing and mode identifying ofthe Zernike moment can be promoted, the method can be applied to video image watermark, shot division, wave front reconstruction in an optical system, and target positioning and recognition detectionin a machine vision system, and the method provides extremely valuable reference for quick calculation of other orthogonal moments.

Description

technical field

[0001] The invention belongs to the technical field of orthogonal moment acceleration, and in particular relates to Zernike moment theory and a GPU-based fast calculation method. Background technique

[0002] Image is a very useful information medium and communication tool, which can express and process complex scenes in a compact and efficient way. Moment invariance theory is widely used in the fields of computer vision, remote sensing processing, medical imaging, pattern recognition and image processing as a feature to describe images. Moment theory originates from mathematics, physics and statistics, and is a scalar used to characterize functions and capture important properties of functions. Moment descriptors are region-based image features because they use all the information of the image, i.e. the image contour and its content. Unlike descriptors based only on contours, such as Fourier descriptors that use image boundary information.

[0003] In sum...

Examples

Embodiment Construction

[0040] The present invention is described below in conjunction with accompanying drawing.

[0041] 1. A hybrid algorithm that stores radial polynomial coefficients combined with image re-layout gossip-limited symmetry algorithms

[0042] 1.1 Store the polar diameter and polar angle values ​​of the first hexagram limit

[0043] figure 1 The mapping transformation of mapping an N×N image coordinate [i, j] to an inscribed unit circle [x, y] in Cartesian coordinates is illustrated. The number in each square represents the address index of the pixel. The normalized coordinates are:

[0044]

[0045] Wherein, i=tid%8, k=tid / 8, tid is a thread index, and its value ranges from 0 to 63. according to figure 2 The read order shown in (b) loads the index values ​​using figure 2 Calculate the corresponding polar radius ρ and polar angle θ for the address index value in the first hexagram limit in (a). Its calculation result is as figure 2 (d), store them in global memory.

...

technical field

[0001] The invention belongs to the technical field of orthogonal moment acceleration, and in particular relates to Zernike moment theory and a GPU-based fast calculation method. Background technique

[0002] Image is a very useful information medium and communication tool, which can express and process complex scenes in a compact and efficient way. Moment invariance theory is widely used in the fields of computer vision, remote sensing processing, medical imaging, pattern recognition and image processing as a feature to describe images. Moment theory originates from mathematics, physics and statistics, and is a scalar used to characterize functions and capture important properties of functions. Moment descriptors are region-based image features because they use all the information of the image, i.e. the image contour and its content. Unlike descriptors based only on contours, such as Fourier descriptors that use image boundary information.

[0003] In sum...

Examples

Embodiment Construction

[0040] The present invention is described below in conjunction with accompanying drawing.

[0041] 1. A hybrid algorithm that stores radial polynomial coefficients combined with image re-layout gossip-limited symmetry algorithms

[0042] 1.1 Store the polar diameter and polar angle values ​​of the first hexagram limit

[0043] figure 1 The mapping transformation of mapping an N×N image coordinate [i, j] to an inscribed unit circle [x, y] in Cartesian coordinates is illustrated. The number in each square represents the address index of the pixel. The normalized coordinates are:

[0044]

[0045] Wherein, i=tid%8, k=tid / 8, tid is a thread index, and its value ranges from 0 to 63. according to figure 2 The read order shown in (b) loads the index values ​​using figure 2 Calculate the corresponding polar radius ρ and polar angle θ for the address index value in the first hexagram limit in (a). Its calculation result is as figure 2 (d), store them in global memory.

...

Images