Salient Object Detection Based On Eye Tracking Data And Multi-scale Convolutional Neural Network

With the rapid development of Internet and multimedia technology,image has become an important carrier of information transmission.It has become a hot issue to effectively filter the redundant information and extract the saliency region from the amount of image data,image salient object detection is an important way to solve the problem.Generally,the low-level features based saliency detection model has achieved good performance,however,it without take into account the human's visual attention and high-level semantic feature,which can't satisfy the demand of complex image.Inspired by the fact that convolutional neural network has forceful ability to extract high-level semantic features from raw pixels automatically,it has been widely utilized in image processing.By recording human eye movement information in watching image,eye tracking data provides a basis for understanding the visual perception and real intention of the human eye,which can be used to simulate people's visual selective attention mechanism.The main works are summarized as follows:(1)A novel salient object detection model based on multi-scale convolutional neural network is proposed.Firstly,we use a recurrent neural network to extract high-level semantic information and exploit deconvolution layers to mapping the information to hierarchical feature maps.Then,multi-scale saliency prior maps can be extracted from based on these different feature maps,which can be fused to predict the last saliency map.(2)We proposed an eye tracking based salient object detection algorithm.Eye tracking saliency map is obtained by fixation data,which can be used as the forth channel of original image to detect salient objects in an image.Experiments show that the eye tracking data can effectively improve the performance of saliency detection model.