| Abstract [eng] |
The usage of multimedia technologies in the television and film industry is growing rapidly each year. The isolation of the area in the video material has become a standard procedure in the digital image processing sequence. Despite the fact, that the market is being realized in the field of different kinds of extraction methods, the need for their analysis and improvement remains. This master thesis is written in order to evaluate the statistical which burden background extraction from video footage. After clarifying these reasons it is allowed to objectively determine the image quality, and take into account their minimization during the development of videos. The first part of the research is featured in the mathematical area of isolation and the color space analysis methods. These two topics are related to each other, whereas the color information is used in the field of a extracting object from video background. In order to achieve master thesis‘s objectives, the second section of the work describes the process of the realization of the extraction methods and the RGB and HSV color spaces. In addition, this section includes the most functional program choice. Statistical parameters were also briefly reviewed, which examined data sets. These videos are being generated with various conditions through motion graphics software. The results obtained were tested with SSIM, 3SSIM, PNSR and MSU Blur statistical estimates. With the help of statistical parameters, it was found that the video extraction methods of different color spaces RGB and HSV generates nearly identical results. At the time of testing the statistical parameters have clarified the most difficult extraction aspects. The greatest influence on the structure of the results had the video motion blur. Intensive effect of this process, the structure of the original file in the isolated area, was the most different. These results clearly revealed during a tests with statistical parameter 3SSIM which compares borders of the objects, textures and gradient smooth areas. |
