Estimation of absorber depth in transillumination image by deep learning

摘要

To estimate the absorber depth in turbid medium from a two-dimensional transillumination image,we developed a new technique using the deep learning principle. The image elements blurred bythe light scattering in turbid medium were generated by the convolution of the original imageelement and the point spread function (PSF) which is a function of the absorber depth. These blurredimages and the given depths were used as training data for the convolutional neural network (CNN)of the deep learning. After the training, we can obtain the absorber depth as the output of the CNNwhen we input a blurred image of unknown depth. In a computer simulation, the validity of theproposed technique was verified. In an experiment the estimated depths agreed well with the givendepths in the range of 2 ~ 6 mm within 14% error.