The Research And Application Of Human Body Contour Extraction With Saliency Detection

Nowadays, the advancement of technology, masses of image and video data is over the network, enriching people’s cultural life, increasing people’s knowledge. In order to mine image information deeply, correctly understand the surrounding environment, quickly capture main content of images, the technology of salient region detection came into being.Salient region is often the most attracting area, which contains most of the information on an image, therefore, detecting the salient region plays an important role in the image analysis and processing. The salient detection technology has broad application background, usually, it can be used in object recognition, image segmentation, image retrieval and other fields, a fast and effective salient detection technology helps the development of these areas a lot.Now, many different kinds of salient detection technology exist, and they are mainly divided into two general direction:local contrast methods and methods based on global contrast. The former methods calculate the significance using the information of the neighborhoods, since these methods only concern with the image information in adjacent areas without considering the overall impact, they can only highlight a part of the salient region. And they perform bad for the images with complicated background. While,the latter methods are based on the overall relationship and structure which significance is defined by the difference between one region and the whole image, thus,they can highlight the entire salient region more accurately, but the calculation is relatively complicated. In order to solve the above problems, better salient detection technologies are needed to promote the development of the image field.Salient detection technologies are not only used for the natural images, they can also be used for medical projects to extract the human body contour. In image information acquisition module of medical projects, it often needs to take pictures, extract the human body contour, and sketch out the affected area for follow-up treatment of the disease. The conventional human body contour is extracted mainly by a region growing method, firstly, you need to set the location of diffusion seeds, then, seeds continue to spread out, increasing the area until no growth any more, lastly, the contour of the salient region is extracted. This method is simple, easy to understand, but it can be affected by people, and the position of the seeds plays a great role in the result. While, it is easy to find that the main subject in the people’s pictures taken in medical projects is people and people is the salient subject in the pictures. Therefore, To compensate for regional growth insufficient in extracting the body contour, the paper comes up with a new idea from the viewpoint of salient region detection. It transforms the problem of body contour extraction to the process of detecting significant image region.After learning and researching the knowledge of image salient region detection, this paper proposes two new and significant image region detection technologies to extract the human body contour:1. A contour extraction algorithm for image salient region based on superpixelsFirstly, we use superpixel generation approach to get over-segmentation of the given image. Then, we use color, significance and gradient to define a new similarity metric for over-segmented regions. Based on the new metric, a region merging algorithm based on the over-segmentation graph is presented. Finally, we can extract the contour of image salient region.2. An automatic contour extraction algorithm for robust image salient regionFirstly, it can calculate the boundary connectivity values by detecting the background, then use SC which is based on H-SLIC to obtain the salient map, lastly, combine the boundary connectivity values and the salient map to complete the segmentation and output the segmentation map.That’s also to say, The contour of the salient region are extracted now.Experimental results show that the two salient region detection methods can extract the contour of the body image well, and they are simple, easy to understand, easy to operate, can be used for medical and other projects to help in treating disease, benefiting thousands of families.