Complex Texture Image Saliency Detection Based On Feature Fusion

Saliency detection is a popular content in digital image processing in recent years.It can play a decisive role in the practical application of image segmentation,adaptive compression,face recognition,image retrieval and so on.With the development of the computational model,more and more saliency detection algorithms are proposed,and the detection speed,accuracy and robustness are improved greatly.At present,although the image saliency detection technology has made great progress,but in the actual application process,the complex background texture,the lower target area contrast and image brightness unevenness will have a significant impact on the saliency detection.Because the typical method is mainly based on the contrast of the color feature as the main judgment basis to calculate the saliency value,the texture feature as the basic features of image is not sufficient,in dealing with images containing complex texture,the accuracy and the recall rate is low of the existing algorithm.Aiming at the problem that the current typical method is less effective than the detection of complex texture image,this paper presents a new algorithm for the complex texture image saliency detection based on feature fusion.The main contents are as follows:Firstly,texture features and color features separation.The image texture is removed by a method based on the total variation model,the complex texture which is easily detected as a saliency region is erased from the image,the interference of the texture to the saliency detection based on the color feature is reduced,the algorithm efficiency and the accuracy,And through the use of SLIC superpixels segmentation to enhance the overall detection process accuracy and reduce the computational complexity.Secondly,Using the existing saliency detection method to detect the removed texture image,saliency map based on texture features is obtained.The texture feature map is extracted by Gabor filter,and the texture feature graph is obtained by using the Gabor filter.The difference of the texture feature between the saliency region and the background region can be obtained by the method of the global contrast.Finally,a fusion method of texture features and color features is designed to obtain a saliency map that contains both texture feature and color feature.This method makes full use of the two basic features of texture and color,and reduces the mutual interference between the two features in the process of calculating the contrast by feature separation,and obtains the human visual attention model,which is simple and efficient.Through the experiment,the evaluation results are evaluated.When the evaluation method is used in the open data set MSRA-10 K,which is commonly used in the field of saliency detection,the proposed algorithm is superior to some representative detection methods.