DeepClean: Integrated Distortion Identification and Algorithm Selection for Rectifying Image Corruptions

TL;DR

This work addresses distortion identification and correction-algorithm selection to boost downstream vision tasks without relying on clean references. It introduces DeepClean, a two-level planning framework that first identifies the latest distortion and then ranks candidate corrections by embedding-based similarity, iterating until the image is rectified. The approach uses a ResNet-50 backbone with distortion-specific heads and adopts SCUNet variants for denoising and gamma correction for exposure, achieving high distortion-identification accuracy and superior object detection performance on COCO compared with baselines. Its plug-and-play design enables dynamic reconfiguration to unseen algorithms, reducing manual tuning and enhancing generalizability across diverse distortion scenarios.

Abstract

Distortion identification and rectification in images and videos is vital for achieving good performance in downstream vision applications. Instead of relying on fixed trial-and-error based image processing pipelines, we propose a two-level sequential planning approach for automated image distortion classification and rectification. At the higher level it detects the class of corruptions present in the input image, if any. The lower level selects a specific algorithm to be applied, from a set of externally provided candidate algorithms. The entire two-level setup runs in the form of a single forward pass during inference and it is to be queried iteratively until the retrieval of the original image. We demonstrate improvements compared to three baselines on the object detection task on COCO image dataset with rich set of distortions. The advantage of our approach is its dynamic reconfiguration, conditioned on the input image and generalisability to unseen candidate algorithms at inference time, since it relies only on the comparison of their output of the image embeddings.

Figures (3)

• Figure 1: Description of three levels of hierarchy to generate a Vision Pipeline.
• Figure 2: Framework to identify image distortion.
• Figure 3: An example of the framework to select the correction algorithm for distortion rectification. Assuming that in the first stage if the identified distortion is an "underexposed image".