Improved Error Reduction and Hybrid Input Output Algorithms for Phase Retrieval by including a Sparse Dictionary Learning-Based Inpainting Method

Figure 5

Final estimated object at a certain iteration. (a) Our proposed method. The red pixels in are taken as missing pixels, which are generated from the activation . (b) Classical ER method. Here, the same initial Fourier phases are given and the activation on the object constraints is selected during PR iteration to fairly examine the performance improvement when applying the image inpainting technique. The estimated object from our proposed method (ER + KSVD) shows that its peak signal-to-noise ratio (PSNR) is higher than using the classical ER method.

(a)

(b)