基于PCA的内容保持图像缩放方法

Content-aware Image Scaling Based on PCA

针对图像的非等比例缩放问题,提出了一种基于主成分分析的内容保持图像缩放方法。将数字图像的列/行看作是图像数据的特征维度,利用PCA算法确定其主成分,定义并计算各列/行的影响度,标注影响度低的列/行,让图像缩放发生在影响度低的列/行上。在标注列/行时采用了影响度扩散的方法,从影响度最低的列/行开始,每标注一列或一行,就将该列/行的影响度扩散到邻近列/行,避免了因连续缩放列/行像素而导致的整体概貌失真。试验结果表明:该缩放方法既能保持图像主体内容,也能兼顾图像整体概貌。

Aiming at the problem of non-uniform image resizing,a content-aware image scaling based on principal component analysis(PCA)is proposed.Taking the column/row of digitalimage as the feature dimension of image data,the PCA algorithm is used to determine its principal component.Define and calculate the influence weight of each column/row,label the column/row with low influence weight,so that image scaling occurs on the column/row with low influence weight.When labeling the column/row,the influence weight diffusion method is adopted.Starting from the column/row with the lowest influence weight,the influence weight of the column/row is transferred to the adjacent column/row to avoid the overall profile distortion caused by the continuous scaling of column/row pixels.The experimental results show that the method can not only keep the main content of the image,but also take into account the overall picture.