The Research And Application On Saliency Detection Of Images Of Fusion Feature

With the rapid development of the Internet and social network, images become more important in information transmission while the problem facing us now is how to process a large number of image information quickly and accurately. Visual saliency estimation can process images appropriately without prior knowledge, and salient area detection can detect significant information in image. Saliency detection is an important procedure in computer vision tasks, such as image transmission, image retrieval, image segmentation, image editing, object recognition and adaptive compression. Therefore, further research in saliency detection has very important significance for image processing, image understanding and future development of mobile Internet.Through the efforts of many researchers, saliency detection has developed into relevant theory, and many saliency detection algorithms have been proposed. However, we are still faced with a series of challenges which can be expressed in two aspects: one is the quality of saliency map is not high enough and can not fully demonstrate the important information of the image. For the image scene of single small saliency objects which is the most common in social network, there are no efficient saliency detection algorithms. On the other hand, the method of application of the saliency information is not good enough and the saliency information is not used efficiently in many application situations.In response to these problems, the research of this paper is from following aspects:(1) Summary the existing research of image saliency detection, including visual saliency model, salient features, principle salient detection, assessment methods, and so on. Then describe some typical image saliency detection algorithm in detail, and compare their characteristics, analyze their application scenarios.(2) Present an improved FT algorithm and a new saliency detection algorithm which integrates various features. This paper improves FT algorithm by increasing the weight of the low frequency part of images. For the image scene of single small saliency objects which is the most common in the social network, the algorithm takes the advantage of the frequency, the color and the position information in such images by combining LC algorithm and FT algorithm, and optimizing location information and post-treatment enhanced the difference of saliency. By contrasting saliency map and precision-recall curve, the algorithm performs excellently.(3) Research the application situations of saliency detection technology. For content-aware background blur, non-photorealistic rendering and bandwidth adaptation of wireless image transmission, this paper uses of existing technology, study the application of the method of saliency information, expands the new application scenarios, and analyzes its feasibility theoretically as well as gets a better experimental results.