基于二进小波变换的边缘保持图像插值算法

Images Interpolation Based on Edges-Preserving Using Dyadic Wavelet

利用离散二进小波变换(DDWT)所具有的良好的多尺度边缘提取特性以及子带间的相关性,提出了基于DDWT的边缘保持图像插值算法。算法利用多尺度边缘的指数衰减规律预测损失掉的高频子带中的边缘,并利用三次样条插值算法恢复损失掉的高频子带中的非边缘信息。最后通过离散二进小波合成得到高分辨率图像。实验结果显示,该算法优于传统的双线性和双三次插值算法。对于像Lena这样的纹理较少的图像其峰值信噪比(PSNR)提高了2dB以上,而对其它图像插值的结果也有不同程度的质量改善,插值的结果更符合人的视觉系统特性。

An image interpolation scheme was introduced based on edges preserving using discrete dyadic wavelet transform (DDWT). By the exponential decay property of the multi-scale edges, the edges of the lost high frequency subband are predicted. And the other coefficients of the high frequency subband are obtained by cubic spline interpolation algorithm. So the high-resolution image may be generated by the wavelet synthesis. Experimental results show that the interpolated images of the algorithm are better than conventional algorithms such as bilinear and bicubic interpolation. For images that own less texture like Lena, the PSNR value increased about 2 dB. And the interpolated image quality of others may also have some increasing. The high-resolution images are more in of accord with human visual system characteristic.