Visual Saliency Detection Research Based On Multiple Domains And Features

Visual selective attention mechanism is inspired by the biological visual system to simulate the visual function of the organism.It is extracted from a complex scene which is easy to cause the attention of the viewer.Visual selective attention mechanism provides a new way to solve the calculation of large amount of data and to improve the real-time perform.Visual saliency detection can take the initiative to select the surrounding environment differences between the larger target to do follow-up treatment.It is equivalent to filter the redundant background information shielding,reducing the processing burden of image analysis system.Significant object detection in image processing tasks is often used as a preprocessing module to provide location information for target detection.It is widely used in the field of target tracking,image segmentation,image compression and so on.It has great application value in the future.On the basis of introducing the human visual perception mechanism,this paper studied most of methods of selective attention mechanism in spatial and frequency domains,and determined the performance of the method which is based on the combination of frequency domain and spatial domain.In this paper,the method is integrated in order to make the local information of the spatial domain and the global information of the frequency domain to be complementary to get the better overall information,and obtain comprehensive and significant regional information.It can improve the accuracy and universality of the visual saliency detection.In order to make the fusion performance better,the method of single space frequency domain is needed to achieve the best performance.So in this paper,the model of spatial and frequency domains are optimized in different degree.On the basis of not affecting the detection accuracy,the detection performance is promoted.In the detection method based on the spatial domain,the multi scale fusion is also carried out on the basis of the existing GBVS method in the feature extraction part.And the space domain method uses the characteristic of the color,brightness,direction to carry on the determination to the chart.The improved method not only improves the detection accuracy,but also saves the time.Finally,saliency maps from two domains using the weighted average fusion method detect the frequency and spatial domain methods based on the combination of theory and information on the formation of complementary advantages.The fusion method in complex background has stronger adaptability and accuracy.What's more,the proposed method makes the detection performance has a lot of ascension.The experimental results show that compared with the single domain of different methods,the significance of this method is better than the single theoretical basis.This method achieves high detection accuracy,strong adaptability,and the application requirements of fast computing time.Compared with single domain model,the new model improved 9.5% reletived to spatial domain model,and improved 6.3% reletived to frequency domain model.In this paper,the method of image segmentation,image background image and other applications have also achieved good results and it shows that our method will have great application prospect in actual image processing in the future.