Image Saliency Detection Technology Based On Superpixel Segmentation

With the advent of the information age,massive digital media has brought great enjoyment to people,and it also puts forward higher requirements for the processing efficiency of digital image processing technology.Especially for the stage of calculating saliency value in the process of saliency detection,the calculation of saliency value at the pixel level will be very large.In order to reduce the operation level in saliency calculation,we need to degrade the original input image by pixel magnitude,so the introduction of superpixel segmentation in image saliency calculation becomes a good and necessary solution,which is also a trend in the future.The main research work of this paper is as follows.Firstly,an image saliency detection algorithm based on spatial distribution and texture features is proposed.This algorithm improves the image saliency detection algorithm based on spatial distribution features.First,this algorithm extracts spatial saliency map using the original algorithm.In order to consider more underlying feature information comprehensively,the algorithm then fuses texture features to get the final saliency map.The experimental results show that the subjective and objective evaluation indexes of the algorithm in this paper's experiments are better than some classical saliency detection algorithms.Secondly,an image saliency detection algorithm based on improved SLIC superpixel segmentation is proposed.In order to avoid the problems that the traditional SLIC superpixel segmentation algorithm brings too many iterations,generates redundant cluster centers and generates fewer superpixels,the algorithm first uses the improved SLIC superpixel algorithm instead of the traditional SLIC superpixel algorithm.Then in the saliency calculation stage,in order to make the result saliency map contain richer feature information and to uniformly highlight the target region of the image while retaining the boundary contour,the algorithm fuses texture saliency map and color histogram-based saliency map to get the final saliency map.The experimental results show that the subjective and objective evaluation indexes of the algorithm in this paper's experiments are better than some classical saliency detection algorithms.Thirdly,an image saliency detection algorithm based on improved Retinex algorithm and improved SLIC superpixel is proposed.In order to improve the applicability of the previous algorithm for images with uneven background illumination,low image quality and environmental impact,the algorithm first improves the traditional Retinex algorithm by particle swarm optimization and applies the improved Retinex algorithm as a preprocessing operation to the image saliency detection process based on improved SLIC superpixel segmentation.In order to further reduce the background noise in the saliency map based on the improved superpixel extracted by the previous algorithm,the algorithm then extracts the foreground saliency map,and fuses the foreground saliency map with the saliency map based on improved superpixel to obtain the salient image based on bottom-up feature information.The experimental results show that the subjective and objective evaluation indexes of the algorithm in this paper's experiments are better than some classical saliency detection algorithms.Fourthly,an image saliency detection algorithm combining local feature information and global feature information is proposed.In order to better highlight the core target area in saliency map,the saliency map based on bottom-up feature information extracted from the previous algorithm is optimized by POO algorithm.In order to give better consideration to global shape information,the algorithm then trains a strong classifier of multi-core functions to optimize the saliency map optimized by POO algorithm,and obtains saliency map based on global feature information.Finally,the bottom-up local feature saliency map and top-down global feature saliency map are fused to get the final saliency map.The experimental results show that the subjective and objective evaluation indexes of the algorithm in this paper's experiments are better than some classical saliency detection algorithms.