局部自适应学习基稀疏约束结合信息优先权选择扩散的迭代图像修复算法研究

Iterative image inpainting using sparse constraint with local adaptive learned dictionary and informational priority selected diffusion

图像修复的性能依赖于局部邻域信息的利用,同时要求修复的方向与结构性内容具有一致性,针对这两个关键问题,本文提出基于局部自适应学习基稀疏约束结合信息优先权选择扩散的迭代滤波图像修复思路。该算法首先学习丢失区邻域得到一组具有局部自适应性的稀疏基,然后沿着等照度线方向按照优先权选择扩散的顺序利用稀疏重构理论以这组基逐层投影重构丢失区域,通过迭代执行分层修复实现对丢失区域的渐进逼近。实验结果表明,该算法无论对于纹理图像、边界图像还是自然图像都可达到较好的效果,而且在峰值信噪比上优于已有文献的方法。

The performance of image inpainting is dependent on the utilization of local neighborhood information and the inpainting direction is supposed to keep consistent with the structure information for avoiding wrong diffusion. Aiming at these two issues,iterative image inpainting using sparse constraint with local adaptive learned dictionary and informational priority selected diffusion is proposed in this paper.Firstly,the algorithm obtains a group of local adaptive sparse bases by training a set of samples selected from the neighborhood of the lost area. Secondly,a hierarchical model is adopted in the step of image inpainting.According to sparse reconstruction theory, the lost area is reconstructed from the projection layer by layer along isophote direction using the local adaptive sparse bases with the priority selected diffusion method.This process is performed through iteration process until the result is convergent.Experiments on texture images,structure images and real images demonstrate that the proposed algorithm is efficient and could achieve better performance.