基于Curvelet的纹理方向自适应图像插值

为提高纹理复杂图像的插值放大质量,提出一种纹理方向自适应的图像插值算法。首先利用Curvelet变换提取图像的4个方向因子矩阵,然后对不同类插值点选择相应的2个方向因子构造权重系数进行线性插值,从而自适应地重建各类待插值点。分析比较了本文与现有插值算法对平滑与纹理复杂区域的插值重建质量。实验结果表明,本文算法能有效抑制传统插值方法重建图像时出现的边缘模糊和锯齿现象,重构的图像效果优于传统方法,纹理丰富图像的重构质量可提高2dB以上。

Texture direction adaptive image interpolation based on Curvelet transform

To improve the interpolation quality of the complex image,an adaptive image interpolation algorithm is proposed based on direction feature of texture components.The four direction factors of an image are estimated using the Curvelet transform.For each interpolated point,the two weight coefficients are computed by the two direction factors chosen by the location of the interpolated point and used to reconstruct the value of the interpolated point by means of linear weighting.Experimental results show that the interpolated image with our algorithm has the superior performance in both the smooth region and texture-rich region.Compared with the traditional methods,the proposed method could remove the jaggy and blur at the edge effectively.The peak signal to noise ratio(PSNR) has also been improved by more than 2 dB for the image which contains lots of texture information.