Fig. 1. Statistical characteristics analysis of the different sub-regions of one speckle pattern. (a) Schematic of speckle pre-processing step and selection of different sub-regions; (b) Speckle of four different sub-regions and corresponding power spectrum and reconstruction results; (c) Intensity value corresponding to the dotted line of the normalized power spectrum pattern

Fig. 2. Schematic of the physics-aware learning method for efficient phase retrieval algorithm. (a) Speckle power spectrum pre-processing module; (b) Neural network post-processing module for object recovery

Fig. 3. Experimental setup for the scalable imaging

Fig. 4. Results comparison without or with physics-prior information for imaging through unknown diffusers

Fig. 5. Scalable imaging results of human faces data through different diffusers

Fig. 6. Results comparison using speckle redundancy theory(Different numbers of sub-region speckles used and every sub-region size have different effect on scalable reconstruction imaging)

Table 1. Reconstruction indicators corresponding to the use of different numbers of sub-region and different sizes of sub-region speckles