Research On Saliency Detection Method Based On Superpixels

Saliency detection is widely used in daily life,industrial and agricultural production,modern logistics,smart city,military and national defense,and plays an important role in image analysis,computer vision and target recognition.There are some defects in the traditional saliency detection method,and the image processing is carried out at the pixel level,which leads to the tedious detection process and need large amount of data,traditional saliency detection method is difficult to extract the saliency objects in the complex background.Superpixels saliency detection can reduce the amount of data processing,simplify the calculation and improve the detection efficiency.For this reason,based on superpixels,we study the saliency detection of static image and dynamic video respectively.The main work of this paper is as follows:Research on the method of static image saliency detection: Aiming at the problem that the method of saliency detection in complex background can not effectively suppress the background and accurately detect the salient target,a multi-scale Bayesian saliency detection method based on superpixels content perception prior is proposed.Firstly,the image is segmented into multi-scale superpixels images,the contrast priori,center position priori and boundary connected background priori of content perception are introduced into each scale to calculate the target saliency on a single scale;secondly,the multiple scales rough saliency map values are taken as a priori probability,According to the color histogram and the convex hull center priori,the observation likelihood probability is calculated,and then the multi-scale Bayesian model is used to obtain the final salient target.Finally,three open data sets,five evaluation indexes and seven existing methods are used for comparative experiments,and the results show that the method in this paper has better performance in the static image seliency detection.Research on dynamic video saliency detection methods: Aiming at the existing methods which mostly use static spatial features,leading to videosaliency detection can not get accurate saliency target,a video saliency detection method based on the fusion of superpixels gradient flow field and cellular automata is proposed.Firstly,SLIC method is used to segment video frames into superpixels.At the superpixels level,optical flow gradient and color gradient are used to generate a spatiotemporal gradient function,and new gradient flow values are obtained from spatiotemporal gradient,which can make full use of the motion information in the video.Secondly,convolution neural network is used to obtain the depth characteristics of superpixels,and the depth characteristics of superpixels are applied to cellular automata,the rough saliency map is automatically updated according to the custom rules,and then the gradient flow field saliency map and the rough saliency map of cellular automata are fused to get the final saliency map.Finally,the comparative experiments on Vi Sal data set use five evaluation indexes and four existing methods,the results show that the method in this paper has advantages in the dynamic video image saliency detection.