Research On Image Saliency Detection Algorithm Based On Information Fusion

Saliency detection is a popular research in computer vision field.The purpose of saliency detection is to simulate the visual information processing of humanbeing and to select the most attractive part of the image for further processing.Saliency detection is widely used on image segment,object detection,image retrieval and other field.With the development of application background,Saliency detection technology is developing either.In this paper,we reviewed the basic theories and algorithms of saliency detection and proposed two saliency detection algorithms.Pattern mining saliency base on local saliency information of image and it could usually get accurate detection result.But some part of saliency region could be depressed.In order to get an accurate intact saliency region.We add large-scale saliency information to pattern mining saliency algorithm and propose the local and global saliency information based saliency detection method.Experiment also show that the method we proposed could get a good result.Neural network based saliency detection algorithms are task-driven method.In order to utilize saliency prior,combine saliency information of different scales and continuously improve saliency detection results.We merge input image and saliency prior,combine output of different convolutional layers in full convolution neural networks and use recursion full convolutional neural networks for saliency detection.Experiment also show that the method we use could get a good result.