Research On Image Salient Object Segmentation Algorithm

In recent years,the machine vision has already been applied in many aspects of our daily life.As one of the most import part of machine vision,image segmentation plays a key role in the accuracy and reliability of engineering applications.Aiming at improving the accuracy of image segmentation,two types of image segmentation algorithms have been proposed by studying and improving extant image salient object segmentation algorithms.1.As extant bottom-up object segmentation algorithm based on LMLC and GMR wrongly highlight backgrounds in complex scene images,a new algorithm of salient object segmentation is proposed in combination with sparse reconstruction and energy equation optimization.Firstly,the input image is decomposed into superpixels within abstracting unnecessary details by using simple linear iteration clustering(SLIC)algorithm.Then,the image boundary superpixels are selected as the background templates,which are used as sparse dictionaries to calculate reconstruction errors and as the initial salient value of superpixels.Finally,objective function constructed by the manifold ranking energy equation of graph theory is used to optimize the initial salient value,and object segmentation is processed after the optimized salient map foreground is enhanced.The proposed algorithm is tested and compared with other algorithms of the same type.The experimental results show that the proposed algorithm is more robust than other existing algorithms in the image salient object segmentation in complex background images,and the background is suppressed more effectively,and the segmentation of salient object is also more accurate.2.The experimental results show that extant top-down deep convolution neural network image salient object segmentation algorithm missed some structural boundary contour and edge detail information,which also reduced image object segmentation accuracy.Aiming at overcoming this defect of deep convolution neural network and improving accuracy of image object segmentation,a image salient object segmentation algorithm based on deep features guidance is proposed.According to the characteristics of different level of the convolution neural network,a skip cross-level feature-guided joint network model from deep convolution layer to low convolution layer is constructed by highlighting the dominant high-level semantic features.In order to improving the quality of structural boundary contour,increasing edge information and reducing noise,simple linear iteration clustering(SLIC)algorithm and the fully connected conditional random field(CRF)are used to optimize the salient map produced by deep convolution neural network,and the image object is segmented after the salient map isoptimized.The proposed algorithm is tested and compared with other algorithms of the same type.Experimental results show that the proposed algorithm improved the output of the convolution neural network and reduces the impact of defects,and the accuracy of image object segmentation is also improved.