Research On Image Sensitivity Detection Of Fusion Multi - Information

With the development of computer technology and the Internet, especially in recent years, the popularity of smart mobile devices, and the rise of the mobile Internet, a great convenience to people’s lives, study and work. Even people in thousands of miles distant from each other can be known in living conditions-to share their lives via the Internet. In the act of sharing pictures occupy majority, because the image is more direct than text, and easy to understand, and conveyed enriched. Image resources on the Internet has exploded. It is a challenging task to storage and retrieval of the increasing large scale image effectively.It is hoped that a computer to like humans intelligent understanding of the image contents and its corresponding processing. With the development of computer vision and image processing technology, as well as for the study of human visual attention mechanism found human visual system is selective attention mechanism. So inspired by this human visual attention mechanism, will gradually be applied to the human visual characteristics of this video, image processing, etc., in other word, only the video or image area that can cause visual attention will be processed, namely Saliency detection. Then other regions was ignored or only a relatively simple process, which can significantly improve the processing speed of computing, at the same time also can save storage space. Thus saliency detection in computer vision and image processing has a wide range of applications, such as target detection and tracking, image segmentation, object recognition, image compression, image retrieval. Based on the image to start the multi-angle feature summarizing the previous experience, based on the integration of a variety of information on graphics research significance testing, mainly to do the following work:(1)Firstly, a brief introduction of the basic knowledge and concepts saliency detection was proposed, including the basic features of the image, such as color, texture, brightness, etc.; as well as some related concepts, such as contrast, a priori knowledge; there baseline data sets and corresponding evaluation standards. Then simple introduction and summary of some common basic idea of classical methods were.(2)Based on the summary of previous experience, a method based on super-pixel image segmentation and integrated global and local contrast was proposed. Firstly, the image is filtered so that the image smooth and uniform; on this basis, edge of the object was determined by the local sliding window. The super-pixel segmentation of the filtered image, Then global and local image contrast and spatial relationship was integrated to the image determine the saliency of each super pixel block. Finally, the above structure fused and filtered to give the boot to optimize the final test results. Experimental results show the effectiveness of this method.(3)Based on the analysis of image information redundancy, the proposed method for detecting significant fusion image information redundancy and image contrast and spatial relationships. Firstly, a dictionary was got by learing, use the dictionary to obtain valid information of each channel in image, then the original image color quantized to reduce the number of colors, simultaneously image was segmented, image color characteristics and spatial distribution was integrated to calculate the saliency of each region. Finally, fuse above results to get the final saliency map based on integration of complementary information. The comparison with the international mainstream method on publicly available data sets MSRA-1000 with accurate artificial mark which is widely used. The experimental results show that our method has a high precision rate, recall rate, F-measure and lower absolute error. Meanwhile, the experiment also confirmed the validity of the proposed method.