一种新的基于稀疏分解的图像放大方法

Novel Image Zooming Method Based on Sparse Decomposition

提出了一对新的冗余离散小波变换(RDWT)和波原子变换(WAT)字典,并将其应用于图像稀疏形态成分分解以获得图像的卡通与纹理成分。并针对卡通和纹理所具有的不同形态学特征,对卡通成分采用具有曲率运动、边缘冲击特性和平滑去噪性能的非线性self-snake模型来放大;对纹理成分采用双三次插值方法来放大,最后通过叠加就可获得放大图像。实验结果表明,这种基于新字典对的稀疏形态成分分解的图像放大方法相比于传统的基于整幅图像的放大方法能够有效地保护小曲率和大梯度,强化图像边缘,保证纹理细节清晰完整。

Two new dictionaries,RDWT and WAT,were proposed in this paper,and we used them to sparsely decompose one image into cartoon component and texture component.Based on the fact that the cartoon and texture in one image have different morphological characteristics,we zoomed the cartoon by self-snake model with the characteristics of curvature motion,edge shock and smooth denoising,and zoomed the texture by bicubic interpolation.Through superposing the zoomed cartoon and texture,the zoomed image will be obtained.The experiment results show,compared with the traditional zooming methods processing the whole image,the new zooming model based on morphological component decomposition has good performance for enhancing edge,protecting small curvature and large gradient,and ensuring the texture clear and completion.