Efficient Image Feature Extraction through Simultaneous Training of Nested U-Nets

In the last few decades, coarse-graining emerged as a tool in various fields to optimize the speed-accuracy complexity of a system and extract prominent features from it. But, in the field of image processing, its application and advancement are not well understood. Moreover, in image processing, the development of optimal neural architecture and its classification are performed on databases with structured and standard images. Here, a computationally efficient CG model is introduced that is very effective in re-scaling and classification of unstructured images. The research and developments in the field of image processing are primarily focused on the development of new tools and network architecture in order to optimize accuracy, whereas the training process remains trivial. With an increase in complexity, the number of layers, and the use of multiple network models in the system, this training process becomes computationally inefficient and costs its accuracy. Here, a unique method is also provided to combine multiple models together as a single network architecture and train them together as a single model rather than doing it separately for every model.