| The target extraction is to separate target from image, which we are interested in. In many fields such as video production, film special effects need to extract the characters in the image which will be used in image synthesis operation. The main operation of target extraction is image segmentation, the research of target extraction can begin with image segmentation. There are various kinds of image segmentation methods, while it can be roughly divided into two classes, the segmentation methods based on the boundary and the segmentation methods based on region. Edge is the area of image gray mutation, it contains main information of the image, and it is one of the main characteristic of image segmentation depends on. The edge of the image reflects main information of the target we interested in such as shape feature, it can greatly reduce the redundant information processing by edge.This article focuses on the study of extracting the figure from the image in the virtual environment. This article has two main research directions. The first one is studying the SUSAN corner detection in the image segmentation algorithm, algorithm in corner detection algorithm. I study the theory and improve the SUSAN algorithm on the basis of combining with traditional Otsu algorithm. The new algorithm has good experiment effect. The second aspect is from an engineering standpoint, combining with practical application and putting forward a rapid method to extract figure from the image under the environment of virtual instruments. It has realized the rapid processing photo function on the computer, and been used successfully in virtual fitting system in the initial figure image processing with the effect of saving a large amount of manpower and time efficiency. This paper’s innovation points are the follows.1. I study the object extraction from the angle of corner and edge as the target feature. On the basis of the study on a variety of improved SUSAN algorithm, following improvement to the traditional SUSAN operator is put forward.(1) Improvement of the operator template. I amend the traditional37pixel template to the ideal circular template. The area weighted way is put forward by45pixel template to simulate ideal circular template. According to the continuity and connectivity of image edge, and in order to strengthen item of adjacent pixels, the inverse distance weighting of each pixel in the template is put forward. It redefines weights of each position pixels in the template during USAN area calculation, changing the singleness weight in the past. We can get a new hybrid weighted template matrix by the calculation of mixed weighted from two aspects which is used to instead of the original37pixels template, and it has better accuracy.(2) Improvement of the USAN principle. It is researching the reason of false dismissal and false detection in traditional SUSAN operator to get three changes of USAN principle. The modified USAN principle has better precision and noise immunity.(3) Improvement of the geometric threshold. According to the geometric meaning and features of the threshold, the improved SUSAN algorithm which is based on Otsu algorithm is put forward.Compared with the original SUSAN algorithm, this algorithm has better resistance to noise and better segmentation effect.2. Combination with the actual virtual fitting system in practical application, and making out the method to extract the figure target from the picture rapidly on the virtual instrument development environment in view of engineering. According to the characteristics of virtual instrument, Mask method and finalization threshold segmentation method which is based on the saturation color extraction model of color image are put forward. After subsequent morphological processing to optimize template and edge contour to realize to pick up the figure characters in the picture. It is a simple, rapidly and effective algorithm. In the actual virtual fitting system it has a good application and performs perfect. |
PDF Full Text Request