Reconstruction of Adversarial Examples to Restore the Performance of a Deep Learning Model: In Computer Vision Domain

In these days, deep learning and computer vision is a much growing field in this modern world of information technology. Deep learning algorithms has gained great success in different fields like image classification, speech recognition, self driving vehicles, decease diagnostic and many more. But when people were enjoying the success of deep learning algorithms then it is found that these learning algorithms are facing serious threats due to adversarial examples attacks. Adversarial examples are inputs like images in computer vision domain, which are slightly changed or perturbed in such a way that, they are human imperceptible but misclassified by a model with high probability, therefore the performance or prediction of a model is badly affected. We present a simple and easy method to reconstruct adversarial examples which are created due to different adversarial attacks and misclassified by a deep learning model. Our reconstructed adversarial examples are correctly classified by a model again with high probability and restore overall prediction of a model. We will also prove that our method is simple, single step and has low computational complexity. Our method is reconstructed all types of adversarial images for correct classification. Therefore, we can say that, our proposed method is universal or transferable. The data sets used for experimental evidence are MNIST, FASHION-MNIST, CIFAR10, and CALTECH-101.