Research On Saliency Detection Methods Based On The Hypergraphs With Adaptive Multi Scale

Saliency detection is an important issue in image processing and computer vision.In this paper,we discuss saliency detection method based on multiscale hypergraphs constructing by adaptive scaling images in each channel.The work is important in the development of image processing and computer vision.At present there are many classical saliency detection methods,but almost all existing methods did not consider the influences of the differences of R,G,B channel value on saliency detection.This paper introduce the differences of R,G,B channel value into saliency detection based on hypergraph modeling and design the saliency detection method based on adaptive scaling images in each channel.The specific studies are as follows:(1)Considering the sensitivity differences of human eyes to RGB colors,we propose adaptive scaling method of pixel value in each channel and get adaptive scaling images.(2)The adaptive scaling method of pixel value in each channel is introduced into the construction of super-vertex and hyperedge.Through combination of adaptive scaling ratios and a set of fixed empirical values,we obtain the multiscale hyperedges and super-vertexes and thus design adaptive multiscale hypergraphs.(3)We apply the adaptive multiscale hypergraphs to the saliency detection and propose a saliency detection method based on adaptive scaling hypergraphs.The proposed method gets the saliency of super-vertexes and hyperedges by the gradient magnitudes and obtains the saliency maps of multiscale hypergraphs.We get the final saliency map after the fusion of multiple saliency maps.We also utilize image segmentation samples to demonstrate the effectiveness of the results.(4)The proposed method is tested on MSRA-1000,SOD,SED,and ImgSal-50 image datasets and compared with six classical saliency detection methods.Numerical experiments show that the proposed method has improved the performance of saliency detection to some extent,especially for images with little difference of R,G,B channel value range.The proposed saliency detection method based on adaptive scaling image in each channel can be applied to many applications,such as image segmentation,object recognition,adaptive image compression and content based image retrieval,which can improve performance and efficiency.