A salient target detection method based on hypercomplex Fourier transform and mean shift

Abstract

The invention discloses a significant target detection method based on a hypercomplex Fourier transform and a mean shift, comprising the following steps: S1, inputting a hypercomplex matrix, defininga hypercomplex matrix; S2, inputting an image and defining a hypercomplex representation of the image; S3, a spectrum scale space being proposed to process the amplitude spectrum of different scales,and a single parameter smooth spectrum series is generated, which is parameterized by the scale of Gaussian kernel; S4, according to a group of spectral scale spaces obtained by the multi-scale frequency domain waves, and the original phase spectrum and the Euler spectrum, salience maps of different scales being calculated; S5, image segmentation being performed by mean shift; S6, a threshold selection method being proposed, in which a threshold value (Ta) is determined to be twice the average significance of a given image; S7, color reproduction of the salient target. The invention can eliminate the repetitive complex background, effectively suppress the complex background, and accurately locate the prominent target.

Description

technical field [0001] The invention relates to the technical field of salient target detection, in particular to a salient target detection method based on hypercomplex Fourier transform and mean shift. Background technique [0002] In the context of the Internet age, big data has become the mainstream of research. As a participant and builder in the Internet age, people use vision as the main source of information. According to statistics, about 108-109 bits of visual information enter the human eye per second [1]. Such a large amount of visual information requires more Image data to save and transfer. How to use computers to solve the limitations faced by the human brain in processing massive image data has given people more room for imagination, so the discipline of computer vision and image processing emerged as the times require. [0003] Salient object detection is often used as a preprocessing module for image analysis and understanding, and has become a research h...

AI Technical Summary

Problems solved by technology

Aiming at the common problems of low accuracy rate and blurred target boundary in the obtained saliency map in the traditional saliency target detection method under complex background interference, the present invention proposes a saliency detection method based on hypercomplex Fourier transform and mean shift. Sexual Target Detection Methods

Images