Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Progressive Contrast Registration for High-Fidelity Bidirectional Photoacoustic Microscopy Alignment

Jiahao Qin

Abstract

High-speed optical-resolution photoacoustic microscopy (OR-PAM) with bidirectional raster scanning doubles imaging speed but introduces coupled domain shift and geometric misalignment between forward and backward scan lines. Existing methods, constrained by brightness constancy assumptions, achieve limited alignment quality (NCC~$\leq 0.96$). We propose PCReg-Net, a progressive contrast-guided registration framework that performs coarse-to-fine alignment through four lightweight modules: (1)~a registration U-Net for coarse alignment, (2)~a reference feature extractor capturing multi-scale structural cues, (3)~a contrast module that identifies residual misalignment by comparing coarse-registered and reference features, and (4)~a refinement U-Net with feature injection for high-fidelity output. We further propose the Temporal NCC (TNCC) and Temporal NCC Gap (TNCG) for reference-free evaluation of inter-frame temporal consistency. On OR-PAM-Reg-4K (432 test samples), PCReg-Net achieves NCC of 0.983, SSIM of 0.982, and PSNR of 46.96 dB, surpassing the state-of-the-art by over 14 dB at real-time speed. Code is available at https://github.com/JiahaoQin/PCReg-Net

Progressive Contrast Registration for High-Fidelity Bidirectional Photoacoustic Microscopy Alignment

Abstract

High-speed optical-resolution photoacoustic microscopy (OR-PAM) with bidirectional raster scanning doubles imaging speed but introduces coupled domain shift and geometric misalignment between forward and backward scan lines. Existing methods, constrained by brightness constancy assumptions, achieve limited alignment quality (NCC~). We propose PCReg-Net, a progressive contrast-guided registration framework that performs coarse-to-fine alignment through four lightweight modules: (1)~a registration U-Net for coarse alignment, (2)~a reference feature extractor capturing multi-scale structural cues, (3)~a contrast module that identifies residual misalignment by comparing coarse-registered and reference features, and (4)~a refinement U-Net with feature injection for high-fidelity output. We further propose the Temporal NCC (TNCC) and Temporal NCC Gap (TNCG) for reference-free evaluation of inter-frame temporal consistency. On OR-PAM-Reg-4K (432 test samples), PCReg-Net achieves NCC of 0.983, SSIM of 0.982, and PSNR of 46.96 dB, surpassing the state-of-the-art by over 14 dB at real-time speed. Code is available at https://github.com/JiahaoQin/PCReg-Net
Paper Structure (22 sections, 6 equations, 3 figures, 3 tables)

This paper contains 22 sections, 6 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Overview
  4. Registration U-Net
  5. Reference Feature Extraction
  6. Multi-Scale Contrast Module
  7. Refinement U-Net with Feature Injection
  8. Loss Function
  9. Experiments
  10. Datasets and Implementation Details
  11. Datasets.
  12. Implementation.
  13. Evaluation Metrics.
  14. Comparison with State-of-the-Art
  15. Performance analysis.
  16. ...and 7 more sections

Figures (3)

  • Figure 1: Overview of the progressive contrast registration framework. The moving image $I_m$ is first coarsely aligned by the Registration U-Net, producing $\hat{I}^{(c)}$ and multi-scale features. A Reference Feature Extractor captures structural cues from the fixed image $I_f$. The Multi-Scale Contrast Module compares features at four scales (32, 64, 128, 256 channels), generating contrast signals that highlight residual misalignment. The Refinement U-Net with feature injection produces the final registered output $\hat{I}^{(r)}$.
  • Figure 2: Qualitative comparison on a representative 26K test sample. Top row: merged full frame (odd + registered even interleaved) in hot colormap; smooth appearance indicates good alignment. Bottom row: magenta/green overlay where odd columns are magenta and registered even columns are green; well-aligned regions appear gray, misaligned regions show colored fringes. Traditional methods (SIFT, Demons, OF, SyN) and deformation-based methods (VM, TM) exhibit visible artifacts, while our method achieves near-perfect alignment with no visible fringes.
  • Figure 3: Ablation study results showing NCC and SSIM across five configurations. The full model achieves the best performance, with each component contributing to the final result.