期刊名称：International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
印刷版ISSN：2278-1323
出版年度：2014
卷号：3
期号：4
页码：1083-1086
出版社：Shri Pannalal Research Institute of Technolgy
摘要：The existing state-of-the-art edge-aware processing techniques may not be capable to completely preserve its edges while manipulating images. High dimensional filters, most remarkably the local laplacian filters are significant tools for numerous computer graphics and vision tasks. These filters are useful for edge aware processing of images and videos. In recent days, a number of techniques for accelerating edge aware manipulation have been developed by exploiting the separability of these Gaussian filters. In this paper we present a new method for edge-aware video processing, which reduces all the un-necessary artifacts in the image edges. This technique also helps to prevent blurring or distortion of its edges during manipulation. It is based on a key observation that temporally adjacent video frames share, identical visual and semantic contents. Using this prior information, temporal filtering of video frames results in splitting of video volumes to small segments suitable for processing. To correctly classify the motion of each scene point an optical vector flow representation is used. Global optimization is used to join the results of local filters at each step. Different resolution images are constructed using Gaussian pyramid concept. Hence applying mixed domain idea makes the result more accurate. This method solves compound edge aware manipulation tasks of videos.
