Research On Image Saliency Algorithm Based On Feature Fusion

With the rapid development of the Internet multimedia technology,the demand of the public to acquire and share pictures also increases sharply,which makes the computer to automatically identify,understand and analyze the content of pictures become a key technology in the field of computer vision.Saliency detection can quickly identify the region of interest in the image and provide simple and effective content information for the computer.This technology is an important part of the computer vision tasks such as image classification,retrieval,segmentation and target recognition.Despite decades of research and development,there are many excellent algorithms for significance detection.In real life,however faced with the huge number of image scale,complicated image content as well as the practical application of certain scenarios.how to accurately and quickly and effectively detect the interested region,accurate assessment of the testing results and using the significance test methods to solve practical application is still in this area is needed to solve the problem.At present the existing methods of precision and recall rate is relatively low,the F-measure value is low,MAE is higher,in the face of a complex scenario cannot accurately detect the target and background region information significantly more problems,this paper aimed at these problems,using the frequency of the image,color,space contrast three characteristics,through the multi-level integration of the three characteristics of cellular automata to significant testing framework method to detect significant goals,on multiple databases of the experimental results show that the method in the subjective and objective aspects are improved to a certain extent.Specific research contents and major contributions are as follows:(1)The specific research content and main contributions are as follows:the principle of significance detection algorithm,the advantages and disadvantages of existing classical significance detection algorithm,and the data set and evaluation index to be adopted in this paper are clarified.(2)The frequency domain features of the image are calculated.The classical FT algorithm is improved by introducing power law transformation and character normalization,so as to improve the frequency domain features of the image and save the target area information while suppressing the background area information.(3)A new fusion mechanism is proposed based on the features of color and spatial contrast of the image,and the significance of the image is calculated by using multilayer cellular automata to fuse the features of frequency domain,color and spatial contrast of the image.The experimental results are compared and analyzed with those of other classical algorithms,and the advantages of the proposed algorithm are proved.