Image Saliency Detection Method Based On A Pair Of Feature Maps And Its Application

Human visual system has a capability of searching for the interesting target quickly in complex surroundings of people's life,which is the most representative and most salient and most important of the surroundings,that is salient object.Compared with humans,it is a challenging technology for the computer visual detection.In order to make it in various fields get a good application,a large number of algorithms of saliency detection have appeared.Most of the existing models got the incomplete salient regions and background of high salient value.How to improve the accuracy of salient object and obtain a saliency map of high recall rate is the focus of this paper.Human visual system observes the environment both in bottom-up manner and top-down manner.Since most saliency detection is goalless and indefinite,this paper mainly studies bottom-up salient object detection.By describing and researching the existing saliency detection models,a saliency detection method based on a pair of feature maps is proposed and successfully applied to the technique of automatically generate mosaic in the salient region of the input image.The main work and contribution of this paper can be summarized as follows: 1.In this paper,a novel saliency detection method based on a pair of feature maps is proposed,which can combine the features of color contrast and the spatial distribution of image color,and avoid the limitation of the single color difference detection method.Firstly,the input image is pre-processed by SLIC algorithm,and color contrast map is obtained according to the color difference between the pixel blocks.Secondly,K-Means clustering of the image according to the color feature is performed,and initial color space distribution of each class is computed based on the compactness of the spatial distribution and the uniformity of color distribution.In order to avoid the lack of spatial information in the clustering results,the spatial color distribution of each class is mapped to the super-pixel blocks,and the color space distribution map is further optimized.Finally,combine color contrast map with feature map of the image color spatial distribution to obtain the final saliency map.In order to validate the effectiveness of the proposed algorithm,a comparison experiment with several popular saliency detection algorithms is carried out based on the public image test database MSRA-1000.The experimental results demonstrate the effectiveness of the algorithm obtained in this paper.2.To meet the need to automatically generate mosaic in salient region,we propose a general framework to automatically generate mosaic in the salient region of the input image,and saliency detection is applied to the mosaic image.Mosaic is a non-photorealistic rendering technology of mosaic images made by combining tessellations of different sizes,shapes and colors.The purpose of this method is to keep the other regions of the image the same,to generate the mosaic effect in the salient areas so as to realize the privacy protection of the important information.Experimental results show the effectiveness of our algorithm,and the boundary of the mosaic area in the generated mosaic image captures the shape of salient regions.The proposed framework can be used for the image privacy information protection and artistic applications.