高级检索
    李志丹, 和红杰, 尹忠科, 陈帆. 结合颜色和梯度信息的稀疏图像修复算法[J]. 计算机研究与发展, 2014, 51(9): 2081-2093. DOI: 10.7544/issn1000-1239.2014.20130071
    引用本文: 李志丹, 和红杰, 尹忠科, 陈帆. 结合颜色和梯度信息的稀疏图像修复算法[J]. 计算机研究与发展, 2014, 51(9): 2081-2093. DOI: 10.7544/issn1000-1239.2014.20130071
    Li Zhidan, He Hongjie, Yin Zhongke, Chen Fan. A Sparsity Image Inpainting Algorithm Combining Color with Gradient Information[J]. Journal of Computer Research and Development, 2014, 51(9): 2081-2093. DOI: 10.7544/issn1000-1239.2014.20130071
    Citation: Li Zhidan, He Hongjie, Yin Zhongke, Chen Fan. A Sparsity Image Inpainting Algorithm Combining Color with Gradient Information[J]. Journal of Computer Research and Development, 2014, 51(9): 2081-2093. DOI: 10.7544/issn1000-1239.2014.20130071

    结合颜色和梯度信息的稀疏图像修复算法

    A Sparsity Image Inpainting Algorithm Combining Color with Gradient Information

    • 摘要: 现有基于稀疏性的图像修复算法仅利用颜色信息衡量样本块的相似度,易降低修复区域内结构部分的连通性及与邻域信息的连续一致性,同时在全局范围内搜索匹配块也增加了算法的运行时间.为解决上述问题,利用颜色与梯度模值信息度量样本块之间的距离,构造新的相似度以确定块结构稀疏度函数,利用块结构稀疏度确定填充顺序,同时构造新的匹配准则函数寻找匹配块;并利用块结构稀疏度值能够较好地反映样本块所处区域特征的特性,根据块结构稀疏度值自适应确定局部搜索区域大小.并通过实验验证在不同图像中颜色信息与梯度信息所占比例不同.实验结果表明,该算法较对比算法能够更好地保持结构部分的连贯性及与邻域信息的连续一致性,在峰值信噪比上至少提高1dB,并且算法速度提高4~11倍.

       

      Abstract: In the existing sparsity-based image inpainting algorithms, only color information is used to measure the similarity between the exemplar to be filled and other exemplars, and the match patches of the exemplar to be filled are searched in the whole image. These decrease the structure connectivity and the neighborhood consistence of texture region, and increase the time complexity of these algorithm. To address these problems, the color-gradient distance between two exemplars is defined by the color and gradient norm information of them. Using the color-gradient distance, the new patch structure sparsity is constructed to determine the filling order, and the new match criterion is obtained to find the most similar patch. Furthermore, the size of local search region is adaptively decided by the patch structure sparsity values to decrease the time complexity of this algorithm. Also the weighting coefficients of color information and gradient information are different in different images, which is verified through experiments. Experimental results demonstrate that the proposed method has better ability to maintain the structure connectivity and the neighborhood consistence of texture area. The PSNR of repaired image by the proposed scheme is 1dB higher than that by the existing algorithms. Additionally, the speed of the proposed scheme is about 4 to 11 times of that of the existing algorithms.

       

    /

    返回文章
    返回