Saliency Detection Based On Multi-scale Fuzzy Width Learning

In the era of big data,social images and video data can be seen everywhere,and the amount is growing rapidly.It is very important to extract useful information from this large amount of data.Visual saliency(known as saliency detection)is one of the classical ways to find regions of interest in the images,and it has always been a research hotspot in the field of computer vision.In this paper,a new saliency detection method based on multi-scale segmentation and fuzzy broad learning is proposed.The research work of this paper is as follows:(1)First,a simple linear iterative clustering(SLIC)algorithm is used to segment the image into super-pixel blocks of different scales according to three scales.In this way,the color information of the image can be utilized and the integrity of the image content can be well preserved.Then,these super-pixel blocks of segmented images are used to calculate the average color of each block to get the color features,and the local binary pattern(LBP)is used to extract the texture features of these blocks to prepare for the following network training.(2)Then,aiming at the problem that the deep learning neural network used to extract the high-level information of image takes too long,this paper proposes a scheme that uses the fuzzy broad learning system(FBLS)to extract the high-level saliency information of image.By training and adjusting the parameters of the network,the network can generate different saliency maps for the images segmented by three different scales.This method can greatly reduce the time of high-level information extraction.(3)Finally,the initial saliency map is obtained by fusing three different scale saliency maps,and the label propagation algorithm(LPA)is used to further optimize it,and the final saliency map is obtained by enhancing the contrast of image foreground and background.In this paper,a large number of experiments are carried out on six commonly used benchmark datasets.Experimental results show that the algorithm has good performance and high training efficiency.