Information distribution on regions of speckle patterns for imaging of multimode fiber

Multimode fibers (MMF) have been extensively investigated for transmitting images. The transmitting images are distorted into speckle patterns by MMFs, which can be reconstructed by neural networks. We studied the information distribution of MMF speckle patterns for image reconstruction. The speckle patterns, segmented by three methods of segmentation, as Centering (1), Quartering (2) and Surrounding (3), are reconstructed into input images by Complex Artificial Neural Network (CANN). Experimental results show that only about one third of full speckle patterns is enough to reconstruct the original images. The quality of reconstructed image is related to the cropping method with different frequency components in speckle patterns, under the same cropped size, Centering segmentation has 4% performance improvement compared to Surrounding segmentation. Optimized segmentation will improve the quality of reconstructed images.