计算机辅助设计与图形学学报
計算機輔助設計與圖形學學報
계산궤보조설계여도형학학보
JOURNAL OF COMPUTER-AIDED DESIGN & COMPUTER GRAPHICS
2015年
5期
864-872
,共9页
康佳伦%唐向宏%张东%屠雅丽
康佳倫%唐嚮宏%張東%屠雅麗
강가륜%당향굉%장동%도아려
特征分类%局部方差%Mean-Shift%K-SVD%分类稀疏表示%结构稀疏
特徵分類%跼部方差%Mean-Shift%K-SVD%分類稀疏錶示%結構稀疏
특정분류%국부방차%Mean-Shift%K-SVD%분류희소표시%결구희소
characteristic classification%local variance%Mean-Shift%K-SVD%classified sparse representation%patch sparsity propagation
针对样本图像字典自适应性差、有效信息单一、造成图像稀疏表示模糊的不足的问题, 提出一种基于特征分类学习字典的结构稀疏传播图像修复方法. 首先将图像块按特征分类, 根据不同特征的图像样本进行样本训练得到相对应的过完备字典; 然后对不同特征的待修复图像块提取不同的有效信息进行稀疏编码, 使得稀疏表示具有较强的自适应能力; 最后针对结构稀疏传播模型带来的偏差进行修改, 完善结构稀疏的传播机制. 仿真实验结果表明,该方法可以有效地修复图像结构边缘、不规则纹理和平滑部分的图像信息, 修复后的图像质量有较大的提升.
針對樣本圖像字典自適應性差、有效信息單一、造成圖像稀疏錶示模糊的不足的問題, 提齣一種基于特徵分類學習字典的結構稀疏傳播圖像脩複方法. 首先將圖像塊按特徵分類, 根據不同特徵的圖像樣本進行樣本訓練得到相對應的過完備字典; 然後對不同特徵的待脩複圖像塊提取不同的有效信息進行稀疏編碼, 使得稀疏錶示具有較彊的自適應能力; 最後針對結構稀疏傳播模型帶來的偏差進行脩改, 完善結構稀疏的傳播機製. 倣真實驗結果錶明,該方法可以有效地脩複圖像結構邊緣、不規則紋理和平滑部分的圖像信息, 脩複後的圖像質量有較大的提升.
침대양본도상자전자괄응성차、유효신식단일、조성도상희소표시모호적불족적문제, 제출일충기우특정분류학습자전적결구희소전파도상수복방법. 수선장도상괴안특정분류, 근거불동특정적도상양본진행양본훈련득도상대응적과완비자전; 연후대불동특정적대수복도상괴제취불동적유효신식진행희소편마, 사득희소표시구유교강적자괄응능력; 최후침대결구희소전파모형대래적편차진행수개, 완선결구희소적전파궤제. 방진실험결과표명,해방법가이유효지수복도상결구변연、불규칙문리화평활부분적도상신식, 수복후적도상질량유교대적제승.
Sample image dictionary has poor adaptability and simplex valid information, which results in bad image sparse representation. Because of the shortage, this paper discusses a new image inpainting method by characteristics classification learning and patch sparsity propagation. The proposed method classified the image patches by their different characteristics firstly, then got the corresponding over-complete dictionary by training the image blocks of different characteristics and extracted different valid information from these blocks for sparse coding, which makes the sparse representation to have stronger adaptive capacity. Finally, the propagation mechanisms can be improved by modifying the patch sparsity propagation model.Experi-ment results show that the proposed method can work on the edge, irregular textures and smooth portion ef-fectively and make the image quality higher.