Image Saliency Detection Algorithm Based On Multi-Feature Fusion And Depth Prior Information

Image saliency detection is one of the most important technologies in image processing,which is widely used in image compression,image segmentation and other tasks.The bottom-up detection algorithm is difficult to obtain a significant area that closes to the attention range of the human eye on some complex images,and there is less real prior information available for the top-down detection algorithm,this paper presents an image saliency detection algorithm based on multi-feature fusion and deep prior information,the main work is as follows:(1)An image saliency detection algorithm based on FLIC fusion of color and texture features is proposed.The saliency maps of color and texture features based on FLIC are obtained in color channel and texture channel respectively,and the final saliency maps are obtained by linear fusion.Compared with seven bottom-up saliency detection algorithms on four open datasets,PR-Curve,MAE and F_?evaluation criteria figures are drawn.The experiments show that the detection algorithm based on multi-feature fusion has better performance,but there is still a promotion in some specific scenarios.(2)In order to improve the performance of multi-feature fusion saliency detection algorithm,an image saliency enhancement model based on depth prior information is proposed to make full use of image information.In this model,the modified VGG16 is used to extract the feature vectors of image pixels and image regions,and a nearest classifier is trained in feature space to determine the attribution relationship among pixels and image regions.The original image and the ground truth are trained as input,and the relationship between the ground truth region feature and the pixel feature is learned through the network.The original image and the initial saliency map that generated by other methods are tested as input,and zero-shot learning method is used to iteratively improve the input saliency map.(3)Furthermore,a saliency detection algorithm based on multi-feature fusion and depth prior information is proposed by combining the FLIC fusion method of color and texture features with the model based on depth prior information.A group of comparative experiment is designed to draw PR-Curve,MAE and F_?evaluation criteria figures.Final results show that this method can achieve better results.