Research On Detection Method Of Salient Object Based On Fourier Transform

Human visual system can rapidly extract the important content of the scene through only focusing on the area of interest in the scene using visual attention and selection mechanism to understand the scene.In the field of computer vision,visual saliency research is carried out with the aid of human visual attention and selection mechanism,and the Saliency Map is used to describe the area of interest.So it is a powerful method to extract the important contents of the scene rapidly.From the comparative analysis of the existing saliency model,this thesis mainly uses the strategy of the combining of the Fourier phase spectrum method for scene target detection and graph segmentation method for extracting salient objects to research the saliency detection method of static images and video.Moreover,the salient objects in images and videos were extracted using the saliency map.The main research content is as follows:1)Comparative analysis of the significant detection performance of the existing Itti model and IG model for the follow-up research of saliency model;2)Fully analysis of the existing Fourier transform,and study the basic theory of the Fourier transform to extend to four yuan;3)Applied quaternion Fourier transform to the saliency detection of static image.Considering the 2D image brightness and two color channels as3-tubal Fourier transform,extract the salient object by changing the expression of quaternion,and finally analyzing the correlation results of the static image saliency detection and target highlight.4)The quaternion Fourier transforms is extended to the dynamic video to detect and highlight the salient object.And the experimental comparative analysis is implemented by its significance in the dynamic video.Specifically,each pixel of the image is represented by the four elements,such as the brightness,color,and motion.Then,the salient object will be extended.Due to the added motion dimension,this method takes into account the time characteristics between the frame and the frame.Gray and color video experimental results show that the salient object extraction by the quaternion Fourier transform can better overcome the motion and complex background.The saliency models have good stability and practicality,and get satisfactory results.