图像适应算法中非冗余显著图的计算

近些年为了将大尺寸数字图像的内容自适应地显示在手机、PDA（掌上电脑）等屏幕较小的显示终端上以得到更好的显示效果，图像适应得到了很大的关注。现有的大量方法使用了不同的重要度映射来得到最优的图像缩放结果。图像适应算法就是为了去除图像中的不重要信息，为此提出在图像适应方法中使用非冗余显著图的计算来分析不重要信息的位置。首先，引入了图像分解，提出图像高梯度区域的冗余信息。然后，提出纹理梯度的概念，并进一步针对细节信息建立了纹理冗余度分析机制。纹理梯度非常适合卡通图像分析，从而将现有图像适应方法推广到非真实图像。大量实验结果显示，本文方法并不受图像适应算子的影响，该方法对具有复杂背景和包含多目标物体等的图像适应问题都比较鲁棒。

Non-redundancy saliency map for image retargeting

In the past several years, for displaying high resolution images on a limited screen, image retargeting has obtained more attention in the academic and industrial communities. Plenty of previous work used different importance maps to acquire optimal image resizing results. In order to realize the objective, which is to dispose the valueless information, of retargeting,we employ an image information redundancy analysis to guide the resizing process. First, we introduce an image decomposition framework to separate the high gradient redundant regions. Thereafter, we introduce a texture gradient to analyze the texture redundancy. Lastly, we show that the proposed texture gradient is also suitable not only for cartoon image analysis, but also for applying existing retargeting process to deal with non-photo realistic photos. Experimental results show that insensitively to the retargeting framework, our method is quite robust of the images with complex scenes or of multiple objects.