Method, computer program and device for evaluating the robustness of a neural network against image disturbances

Abstract

The invention relates to a method () for evaluating the robustness of a neural network () used for image processing against a group of at least two different disturbances that can be found in images. The method comprises determining a robustness score (SRI), called individual robustness score, of said neural network () for each disturbance (P) in said group; determining at least one similarity score (SSI), called individual similarity score, for the similarity between two disturbances in said group; and calculating a robustness score (SRG), called overall robustness score, of said neural network () against said group of disturbances as a function of: the individual robustness scores (SRI) for all the disturbances (P), and at least one individual similarity score (SSI) for the similarity between two disturbances. The invention also relates to a computer program and a device implementing such a method.