Rosa Image Subtraction Based New Image Background Subtraction Algorithm for Improving Software Reliability

摘要

The present researche proposes a new background subtraction algorithm (Rolling Sphere Algorithm ROSA) intended to implement ?background subtraction using spherical coordinate system based sphere rolling which is a vector application technology for separation between objects and backgrounds in digital images. This ROSA can be largely divided into a pre-treatment stage where target images are processed for the sphere rolling and a stage where the sphere is rolled on the pre-treated target image to subtract the actual background. This algorithm groups the edges of the subject of extraction in the image to divide the image into areas, processes the areas as if they are 3D images by making the areas into layers and make a sphere roll over the layers to subtract the backgrounds. Experiments were conducted to compare the ROSA with the Codebook (CD) and the Gaussian Mixture Model (GMM). The total numbers of successful or failed pixels of backgrounds and objects were counted. In terms of pixel removal accuracy, results that were improved by 15% compared to the CD and the GMM. Since, a process to make the target image into a reduced or compressed image filled with average values between adjacent pixels is necessary and vector sphere rolling is applied on the compressed image to apply the ROSA, it could be seen that flexibility in image magnification/reduction and the accuracy of background subtraction are improved compared to existing algorithms.