Saliency-driven Shape Preservation for Image Resizing
Dong Wang, Xuhong Tian, Yun Liang, Xinchun Qu · 2010
We present a novel image resizing method with mesh deformation technique which ensures important regions undergo a similarity transformation. To accomplish this, a mesh is defined over an image. In initial mesh the similarity invariance of local region is calculated based on triangle similarity. When resizing the image,we construct a quadratic energy to measure the similarity error for each local region.The resizing result is obtained by minimizing the energy function sum with salience as weight. The method is enough efficient for it only solves a system of linear equations.Compared to previous methods, our method can preserve the shape of important region better.