Evaluation of image deblurring methods via a classification metric

摘要: The performance of single image deblurring algorithms is typically evaluated via a certain discrepancy measure between the reconstructed and ideal sharp image. choice metric, however, has been source debate also led to alternative metrics based on human visual perception. While fixed may fail capture some small but visible artifacts, perception-based favor reconstructions with artifacts that are visually pleasant. To overcome these limitations, we propose assess quality images task-driven metric. In this paper consider object classification as task therefore use rate metric performance. our evaluation data different types blur in two cases: Optical Character Recognition (OCR), where goal recognise characters black white image, no restrictions pose, illumination orientation. Finally, show how off-the-shelf benefit from working deblurred images.