基于图像分解的含噪破损图像修复

SID Image Decomposition Based Noise Image Inpainting

对Meyer图像分解模型进行改进,所得到的新模型SID(Simplified image decomposition)无需Banach空间的范数计算,降低了计算复杂度。使用SID首先分离出图像中的大部分噪声,然后将剩余部分分解成结构部分和纹理部分。对结构部分使用简化的整体变分修复算法,对噪声部分和纹理部分使用基于样本的纹理合成,再将结果融合。实验结果表明,该方法克服了单独使用整体变分修复算法纹理合成的局限性,对噪声具有较好的鲁棒性。

The Meyer image decomposition model is improved. The new model, called the simplified image decomposition （SID） avoids the norm operation in banach space, thus reducing the computing complexity. Firstly, most noise is separated from the image. Then, the image is decomposed into the structure part and the texture part. Simplified total variation inpainting algorithms are used in the structure part, and the pattern-based texture synthesis in the noise part and the texture part. Finally, results are fused. Experimental results prove that the method overcomes the limitation for using total variation inpainting algorithms or the texture synthesis and is more robust in the noise.