非负稀疏编码收缩法的自然图像消噪

Natural image denoising method based on nonnegative sparse coding shrinkage

非负稀疏编码(NNSC)算法仅依赖自然图像数据的统计特性,具有自适应性.利用NNSC算法可以成功地提取自然图像的特征基向量;作为对特征基的一个实际应用,提出了一种新颖的用非负稀疏编码收缩技术消除自然图像中的高斯加性噪声的方法.实验表明,提取的特征基向量在时域和频域上都有方向性和局部性,表现了输入自然图像的边缘特性;而且与独立分量分析(ICA)法相比,NNSC法提取的特征基有更清晰的边缘特征.目视效果和归一化信噪比证明了NNSC收缩法的消噪效果要优于稀疏编码(或ICA)收缩法、小波收缩法和Wiener滤波等方法.

Non-negative sparse coding（NNSC） relies on natural image statistical properties, therefore, it is self-adaptive. The feature basis vectors of natural images can be successfully extracted by NNSC. As a practical application of such basis vectors, a novel image denoising method based on NNSC shrinkage technique to reduce Gaussian additive noise was proposed. Experimental results show that the feature bases extracted are localized and oriented in the temporal and frequency domains, and this case shows efficiently edge features of natural images. Moreover, compared with the method of independent component analysis （ICA）, the feature bases extracted by the NNSC algorithm exhibit much clearer edge features. In view of to the visual effect and the normalized signal-ratio-noise （NSNR） values of denoised images, the denoising results show that the NNSC shrinkage method outperforms any other types of denoising methods such as sparse coding （or ICA） shrinkage, wavelet-based soft shrinkage, Wiener filter, etc..