一种新颖的图像多尺度几何变换及其应用

提出了一种新颖的图像多尺度几何变换方法,主要由预处理、方向滤波器组和最优方向小波变换等组成.方向滤波器组将预处理后的高频分量分解为多个方向子带,然后每个方向子带执行改进的最优方向小波变换.该变换兼有Bandelet变换和Contourlet变换的多尺度几何分析特性,能更稀疏地表示边缘和纹理特征.分别将EBCOT编码和硬阈值去噪应用到图像变换系数中,从而实现了有效的图像压缩和去噪,很好地保护了图像细节.实验结果表明,对于纹理和边缘丰富的图像,所提出的图像压缩和去噪方法在视觉质量上明显优于基于Bandelet或Contourlet变换的方法,峰值信噪比也提高了0.1 dB以上.

Novel Image Multiscale Geometric Transform and Its Applications

A novel image multiscale geometric transform method is proposed.It is composed of pre-processing,directional filterbank and optimized directional wavelet transform.After pre-processing the image,the high frequency components are decomposed into several directional subbands by directional filterbank,and the modified optimized-directional wavelet transform is performed for each directional subband.The transform has the characteristics of multiscale geometric analysis of both Bandelet and Contourlet transforms,and it represents the edges and texture features more sparsely.The EBCOT coder and hard-threshold denoising are applied to the transform coefficients individually so as to realize efficient image compression and denoising and preserve the image details well.The experiments show that the proposed image compression and denoising method can significantly outperform those that are based on Bandelet or Contourlet transform in visual quality for the images with abundant edges and texture,and the peak signal to noise ratio is also increased by more than 0.1 dB.