Approximation-based Fault Tolerance in Image Processing Applications

Image processing applications exhibit an intrinsic degree of fault tolerance due to i) the redundant nature images, and ii) possible ability consumers application output effectively carry out their task even when it is slightly corrupted. In this scenario classical Duplication with Comparison (DWC) scheme, that rejects images (and requires re-executions) two replicas’ outputs differ in a per-pixel comparison, may be over-conservative. article, we propose novel lightweight tolerant scheme specifically tailored for image applications. The proposed enhances state-of-the-art by: improving DWC by replacing one exact replicas approximated counterpart, allowing distinguish between usable xmlns:xlink="http://www.w3.org/1999/xlink">unusable instead corrupted uncorrupted ones means Convolutional Neural Network-based checker. To tune introduce specific design methodology optimizes both execution time detection capability hardened system. We report results approach on case studies; our proposal achieves average reduction larger than 30% w.r.t. re-execution, less 4% misclassified unusable images.