Research on image transmission mechanism through a multimode fiber based on principal component analysis

Abstract The real-time transmission of images through a multimode fiber (MMF) is still a challenging research work. One method completes image transmission by measuring and controlling the full complex field of the MMF. Another method uses artificial neural networks to train and construct the inverse transformation matrix of the MMF. However, the class of testing samples usually belongs to the same class used for the training. We propose a method that constructs the inverse transformation matrix of the MMF based on principal component analysis (PCA), which can reconstruct the grayscale images of natural scene at high frame rate and high resolution. Moreover, we use principal component analysis and support vector machine (PCA+SVM) to classify speckle patterns, and then images can be reconstructed based on the inverse transformation matrix of corresponding category. The PCA+SVM+PCA (PSP) method shows high reconstruction accuracy with few training samples. The proposed method demonstrates general imaging capability over a MMF lengths of 1 m, training time used to construct the inverse transformation matrix is short and the hardware requirements are low, which provides a feasible method for the MMF endoscope imaging.