图像小波系数的自相似性在图像去噪中的应用

讨论了图像小波系数的四叉树结构及统计规律,指出在四叉树各层节点的小波系数具有自相似性,利用这种自相似性建立混合高斯模型对小波系数进行描述;此外,四叉树层与层间的节点之间具有相关性,利用小波域系数间的自相似性信息和相关性建立隐式马尔科夫树模型(HMT),用于去除图像信号中的高斯白噪声,并利用Lenna图像进行试验仿真;与传统的低通滤波方法相比,该方法在滤除噪声的同时更好地保存了图像的细节和边缘信息.结论表明,隐式马尔科夫树(HMT)模型较精确的体现了图像的特征,具有较好的去噪效果.

The Application of Own Resemblance of Image Wavelet Coefficients in Image Denoising

The quad tree and probability of wavelet coefficients are discussed. The wavelet coefficients of every level nodes in quad tree have resemblance and the mixture Gaussian model which describe the wavelet coefficients is established by using such resemblance. Besides, the nodes of every level in quad tree have relativity, so the HMT built by the resemblance and relativity of wavelet coefficients is used to wipe off the noises in images, and is tested in Lenna. Compared with traditional low-pass filter, the method saves the information of the details and edges of image better while wiping of the noises. The results show that HMT accurately describes the character of image and has better denoising function.