GEOMETRICAL TRANSFORMATION FOR IMAGE RESIZING
Yan Tao, Rynson W.h. Lau, Xie Zhi Feng, Xu Yun, Huang Liusheng, Ma LizhuangAbstract
Image resizing is a process of adapting an image to the target screen size with a different resolution and/or aspect ratio. Although there are a lot of image resizing methods proposed, they either fail to accurately maintain the appearance of prominent objects or distort region/object boundaries. In this paper, we present a novel approach to image resizing, which applies geometrical similarity transformation to prominent objects robustly while preserving straight edges and curves. To do this, we first identify edge structures of prominent objects. As inspired by conformal mapping, which is widely used in surface parameterization, we then apply a geometrical similarity transformation on prominent objects. In addition, since human eyes are very sensitive to straight edges, we try to preserve them strictly and avoid rotating prominent objects using line constraint. Experimental results show that our new image resizing approach is effective.
Read Submission [764]