Abstract: In today’s society where image data are exploding,how to use computer to efficiently acquire and process image information has become an important research topic.Under the inspiration of human visual attention mechanism,researchers have found that when this mechanism is introduced into machine image processing tasks,the efficiency of information extraction can be greatly improved,thus saving more limited computing resources.Visual image saliency detection is to use computers to simulate the human visual attention mechanism to calculate the importance of the information of each part in the image,which has been widely used in image segmentation,video compression,target detection,image indexing and other aspects,and has important research values.This paper summarizes and introduces the research situation of image saliency detection algorithms.Firstly,it takes information-driven sources as starting point to summarize the saliency detection model,and then analyzes several typical saliency detection algorithms.The models are divided into 3 categories according to whether they are based on learning models,which are based on non-learning models,based on traditional machine learning models and based on deep learning models.For the first category,the paper compares in more details the saliency detection algorithms based on local contrast and global contrast,and points out their respective advantages and disadvantages.For the latter two categories,this paper analyzes the application of machine learning algorithms and deep learning in saliency detection.Finally,this paper summarizes and compares the existing saliency detection algorithms and prospects the future development direction of the research in this aspect.

Key words: Salientregion detection, Visual saliency detection, Deep learning, Machine learning, Visual attention mechanism