DCA-LUT: Deep Chromatic Alignment with 5D LUT for Purple Fringing Removal

TL;DR

This work tackles purple fringing caused by Longitudinal Chromatic Aberration by introducing DCA-LUT, a physics-informed, data-driven framework. It first maps RGB into a Chromatic Aberration Space using a learned 3×3 matrix (CA-CT), isolating the fringe into a dedicated channel, then applies a direction-aware 5D-LUT guided by fringe gradients to correct luminance, complemented by a 1D-LUT for residual fringe. The model is trained end-to-end with a composite loss including a perceptual component and a new axis-alignment regularization, and evaluated on PF-Synth and real-world data using standard metrics plus a novel Edge Chromatic Aberration Score (ECAS). Results show state-of-the-art performance in both fidelity and chromatic artifact removal, with superior edge preservation and efficiency suitable for mobile deployment. This approach demonstrates a practical, scalable pathway for physics-informed, spatially-aware LUT-based artifact correction in digital imaging.

Abstract

Purple fringing, a persistent artifact caused by Longitudinal Chromatic Aberration (LCA) in camera lenses, has long degraded the clarity and realism of digital imaging. Traditional solutions rely on complex and expensive apochromatic (APO) lens hardware and the extraction of handcrafted features, ignoring the data-driven approach. To fill this gap, we introduce DCA-LUT, the first deep learning framework for purple fringing removal. Inspired by the physical root of the problem, the spatial misalignment of RGB color channels due to lens dispersion, we introduce a novel Chromatic-Aware Coordinate Transformation (CA-CT) module, learning an image-adaptive color space to decouple and isolate fringing into a dedicated dimension. This targeted separation allows the network to learn a precise ``purple fringe channel", which then guides the accurate restoration of the luminance channel. The final color correction is performed by a learned 5D Look-Up Table (5D LUT), enabling efficient and powerful% non-linear color mapping. To enable robust training and fair evaluation, we constructed a large-scale synthetic purple fringing dataset (PF-Synth). Extensive experiments in synthetic and real-world datasets demonstrate that our method achieves state-of-the-art performance in purple fringing removal.

Figures (9)

• Figure 1: This figure illustrates the physical root of Longitudinal Chromatic Aberration (LCA). (a) Normal Case: All color channels are perfectly aligned, resulting in a sharp, clean image. (b) Case with LCA: Different colors focus at different depths, causing red/blue shift relative to green, resulting in purple fringing.
• Figure 2: Comparison of fringe detection between traditional heuristic methods article and our learned CA-CT module. Our method accurately isolates the fringe artifact while preserving legitimate purple content.
• Figure 3: The overall pipeline of DCA-LUT framework.(a) Chromatic-Aware Transformation (CA-CT). The input image is first processed by a ConvNeXt-based encoder to map into our learned Chromatic Aberration Space, decoupling the image into Luminance Channel ($C_{lum}$), Fringe Channel ($C_{fringe}$), and an Orthogonal Channel. (b) Direction-Aware Correction: A dual-LUT correction branch, where the luminance channel is corrected by Direction-Aware 5D-LUT, utilizing the geometric gradients from the fringe channel to guide the luminance correction. Concurrently, the fringe channel is refined by a 1D-LUT.
• Figure 4: Visual comparison on our PF-Synth dataset. While competing methods often introduce secondary artifacts such as darkening, color casts, or detail loss, our approach is the only one that cleanly removes the purple fringe while faithfully preserving the original image content. (Please zoom in for the best view.)
• Figure 5: Comparison of correction strategies. Green-channel correction (top) introduces a noticeable green tint, while our luminance-based approach (bottom) preserves natural color fidelity.