Using Deep Learning for Object Distance Prediction in Digital Holography

Deep Learning (DL) has marked the beginning of a new era in computer science, particularly in Machine Learning (ML). Nowadays, there are many fields where DL is applied such as speech recognition, automatic navigation systems, image processing, etc [1]. In this paper, a Convolutional Neural Network (CNN), more precisely a CNN built on top of DenseNet169, is proven to be helpful in predicting object distance in computer-generated holographic images. The problem is addressed as a classification problem where 101 classes of images were generated, each class corresponding to a different distance value from the object at a micrometer scale. Experiments show that the proposed network is efficient in this context, being able to classify with a 100% accuracy level if trained properly.