【摘要】 提出一种基于改进支持向量机(SVM)分类和稀疏表示的图像混合去噪算法.首先将输入噪声图像分成大量的重叠片,然后使用尺度不变特征变换(SIFT)从每个片中提取局部特征.根据预定义的阈值,利用粒子群聚类的SVM决策树将贴片分成纹理与平面两类.纹理块利用梯度直方图保存(GHP)进行处理,使用基于差异系数的稀疏度自适应SK-SVD来分析重构平面块.最后,通过合并两个去噪结果获得重建图像.对一些标准噪声图像进行实验,并将本文结果与其他去噪方法进行比较.实验表明,所提出的混合方案具有更好的去噪性能和结构相似性,在保存边缘和纹理方面效果更好.更多还原

【Abstract】 An image hybrid denoising algorithm based on improved support vector machine( SVM) classification and sparse representation is proposed. Firstly the input noise image is divided into a large number of overlapping slices,and then local features are extracted from each slice by Scale Invariant Feature Transform( SIFT). Based on predefined thresholds,the patches are divided into two categories by the SVM decision tree with particle swarm clustering,such as texture blocks and plane blocks. The texture block is processed by gradient histogram preservation( GHP),and the planar block is analyzed and reconstructed by using the sparsity adaptive SK-SVD based on the difference coefficient. Finally,the reconstructed image is obtained by combining the two denoising results. Experiment with some standard noise images and compare the results of this paper with other denoising methods. Experiments show that the proposed hybrid scheme has better denoising performance and structural similarity,and it has better effect in saving edges and textures.更多还原