Saliency Detection Based On Fixation Prediction And Edge Refinement And Its Application

Saliency detection is one of the most important basic research topics in the field of computer vision.In the study of computer vision,the human visual attention mechanism plays an important role in understanding salient objects in images or videos.The visual attention mechanism can enable people to find and select visual targets of interest relatively accurately and quickly in complex scenes.Therefore,how to more effectively simulate the human visual attention mechanism to extract salient objects is a very important research direction of computer vision.With the wide application of deep learning in the field of computer vision,many different methods of saliency detection have emerged.Although the previous method can achieve good remarkability detection results,the detail processing of saliency objects is still not satisfactory.For example,the edges of the saliency target are not refined enough,and the saliency map has a series of defects such as background interference.In summary,this article starts from the essence of saliency detection,and proposes two new saliency detection models to improve the existing problems in the currently proposed methods.The specific research content is as follows:1)This paper proposes an edge refinement network based on eye movement point prediction prior for salient object extraction.Eye movement point prediction can be more in line with the human visual attention mechanism,which is beneficial to quickly and accurately extract the gaze area of the human eye from a complex image background or an image with a blurred foreground.First of all,the input image is predicted by eye movement point,and the resulting feature image is used as a visual a priori for subsequent saliency detection.Secondly,the multi-attention mechanism VGG-16 network is used to extract the salient object features.Finally,the quality optimization process of the feature map is carried out to further improve the quality of the image.The experimental results show that the method of this paper has achieved better significant detection results than the other 6 mainstream methods in 3 open data sets.2)A saliency detection model based on the combination of Res Net and U-Net network is proposed.hen put the processed image into the end-to-end image saliency detection network.The network encoder network uses the Res Net network with the fully connected layer removed,and then uses the multi-scale cascade-channel attention module to extract features,Refinement,and finally the decoding layer is designed as an up-sampling model similar to the coding layer to realize the generation of high-quality saliency images.Experiments on three public data sets show that the method in this paper is better than other representative saliency detection methods in experimental comparison.In short,this article improves and upgrades the existing saliency detection methods,and explores the application of deep learning and visual attention mechanisms to effectively improve the detection of salient objects in images.By constructing two deep learning models,not only has a good extraction effect on the whole,but also has a certain improvement in details such as edges.Experiments prove that the method proposed in this paper can obtain more accurate results than most existing saliency detection methods.