Radon cumulative distribution transform subspace modeling for image classification

We present a new supervised image classification method applicable to a broad\nclass of image deformation models. The method makes use of the previously\ndescribed Radon Cumulative Distribution Transform (R-CDT) for image data, whose\nmathematical properties are exploited to express the image data in a form that\nis more suitable for machine learning. While certain operations such as\ntranslation, scaling, and higher-order transformations are challenging to model\nin native image space, we show the R-CDT can capture some of these variations\nand thus render the associated image classification problems easier to solve.\nThe method -- utilizing a nearest-subspace algorithm in R-CDT space -- is\nsimple to implement, non-iterative, has no hyper-parameters to tune, is\ncomputationally efficient, label efficient, and provides competitive accuracies\nto state-of-the-art neural networks for many types of classification problems.\nIn addition to the test accuracy performances, we show improvements (with\nrespect to neural network-based methods) in terms of computational efficiency\n(it can be implemented without the use of GPUs), number of training samples\nneeded for training, as well as out-of-distribution generalization. The Python\ncode for reproducing our results is available at\nhttps://github.com/rohdelab/rcdt_ns_classifier.\n