应用小波域三维Context模型的视频图像去噪

摘要

考虑视频图像序列的各帧之间具有较强的相关性,提出了一种基于三维小波变换和分块Context模型的视频去噪新方法(3DWTBCM).3DWTBCM法基于视频图像三维小波分解域内系数和噪声分布的特征,利用小波系数具有局部相关性对其进行分块,将系数分解成各个局部区域.然后,将Context模型用于局部块中,按照能量分布将块内的小波系数分成多个子块.对各部分进行能量估计和多阈值估计,获得去噪最佳阈值,并有效地消除噪声.实验结果表明,3DWTBCM的噪声抑制效果明显优于各种2D去噪方法和其他常用的3D去噪声方法,PSNR 平均提高0.5～1.2 dB.而且从视觉效果来看,本文算法在去除噪声的同时,较好地保留了运动图像细节,运动物体显得比较平滑,不存在传统算法中的拖影、闪烁等现象.%A video denoising method based on the 3D Wavelet Transform and Block Context Model(3DWTBCM) is proposed according to the strong correlation between the two frames of video sequence. On the basis of the characteristics of the coefficients in 3D wavelet domain and noise distribution, wavelet coefficients are partitioned into subblocks firstly in the light of local relativity of these coefficients and then the Context model is used in the corresponding subblocks. The wavelet coefficients in each block are divided into several parts by means of their energy distribution in the 3D Context model and each part is estimated by its independent energy distribution.Finally, suitable thresholds are obtained. Experimental results show that 3DWTBCM achieves better denoising performance than hierarchical 2D denoising methods and its PSNR is improved more than 0.5-1.2 dB on average in comparison with those of common 3D denoising methods. In terms of visual quality, 3DWTBCM can effectively preserve the video detail while denoising the wavelet coefficients and especially can provide video frames with rapid movements and more textures.