Research On Salient Region Detection Based On The Superpixel And Texture

There is a part what the observer is interested in the digital images and videos which have a large number of information.It is called to automatically detect these parts and identify the main information quickly in the image.The current saliency detection algorithm employs many different technical structures.With the wide spread of digital images,the performance of traditional salient region detection algorithms cannot meet the growing user needs.Therefore,the research on salient detection algorithm is of great significance.Each perceptual unit can be a pixel or a region.In order to improve the performance of the algorithm,in this paper,we cluster perceptual unit into superpixel,and propose a seed point optimization function.The LBP operator is used to calculate the information of each superpixel,and the superpixel with too much information is divided into two small superpixels.The Improved SLIC algorithm make the generated superpixels more compact and uniform and make sure that the superpixel boundaries adhere well to object boundaries.For simple natural texture images,using a control factor to strengthen the superpixel color contrast,and combined with the feature of space position,analysis the global contrast of each sensing unit,get the final salient map.The concept of textural coarseness is introduced in this paper for texture complex natural images.According to Tamura's visual description of textural and natural images,the coarseness of the whole images is refined to each perceptional unit.Combining color and position characteristics to analyze the saliency of complex texture images.Finally,an adaptive threshold detection is presented.In this paper,the experimental images mainly comes from the MSRA1000 database.Precision,Recall and F-measure are calculating by using ground truth.The algorithm is evaluated and compared with the current classical saliency algorithm.The experimental results show that the algorithm has better precision and can effectively detect the prominent part of the image.In addition,the performance of the algorithm is evaluated on the textural image set which is constructed from the MSRA.This algorithm shows better salient detection performance for the texture complex natural images by comparing the indexes of the algorithm on two databases.