Saliency Detection And Applications Based On Frequency Domain And Spatial Domain

Saliency detection is a process that we use computer to modeling the mechanism of human visual selective attention,and its purpose is to mark the areas of interest of human eyes in the input image.These regions contain most of the information of the image,saliency detection can reduce the computational complexity of other image processing fields for it can inhibit the background noise.and the background interference can be effectively eliminated by saliency detection so as to reduce in other image processing fields.At present,most of the saliency detection algorithms have some problems,such as incomplete significant region,unclear edge,insufficient biological basis and so on.In order to solve these questions this paper propose a novel saliency detection algorithm that utilizes the information of frequency domain and space domain.And more we proposed two saliency-based specific applications.In this paper,we can obtain the prior information by using frequency domain methods,then the foreground prior can be generated by combining with the target prior.We select a superpixels algorithm to segment the image and construct the corresponding similarity matrix.The foreground saliency map is calculated based on the foreground prior information,then we can generate the background seed set with the foreground saliency map.Next,we calculate the contrast of background area and others to generate the background saliency map.Then the initial saliency map is obtained by the fuse of foreground and background saliency maps.Finally,the background saliency map is added to optimize the final saliency map.Experiments on the public datasets of ASD and ECSSD verify that the proposed algorithm can effectively suppress background noise and output relatively complete salient regions.We propose a ship detection method based on the super-complex domain in optical satellite images.This algorithm integrates the color and brightness characteristics of the image into a complex matrix,and the amplitude spectrum and phase spectrum of the complex matrix in the frequency domain.Then we perform wavelet decomposition and reconstruction of the amplitude spectrum to obtain multi-scale saliency maps.Next,we construct an appropriate evaluation function,the final saliency map can be obtained by combining some multiscale saliency maps.Finally,the ship target is obtained by adaptive threshold segmentation.Experiments verify the proposed algorithm can detect ship targets quickly and accurately.This paper proposes an algorithm that multi-scale saliency-based ship detection in SAR Images.Firstly,we employ a threshold to segment the original SAR image and perform a Fourier transform.Then a wavelet can be used to decomposition and reconstruction for the amplitude spectrum coefficients,after it we can obtain the multiscale saliency maps.We defined a function to evaluated the performance of saliency maps and depend on it to select these perform better,these saliency maps can generate the final saliency map.Finally,the ship target can be generated by using an adaptive threshold to split the final saliency map.Experiments on simulated and real SAR images show that the algorithm can significantly enhance the contrast between ships and background noise.