A Linear Fractional Transformation Model and Calibration Method for Light Field Camera

2025-11-06

AI Summary

Low accuracy and poor robustness of light field camera intrinsic calibration hinder its application in 3D reconstruction. To address this, we propose a decoupled calibration model based on linear fractional transformation (LFT), which explicitly separates the geometric relationship between the main lens and the microlens array, while introducing a learnable parameter α to characterize their coupled distortion. We further design a two-stage calibration pipeline guided by analytical solutions: an initial estimate is obtained via least-squares fitting, followed by nonlinear refinement using feature points extracted from raw light field images. Our method significantly improves calibration accuracy and generalization capability, outperforming state-of-the-art approaches on both real and synthetic datasets. Moreover, it enables efficient and photorealistic light field image simulation, providing reliable geometric priors for data-driven light field reconstruction. The implementation is publicly available.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Large Vision Models
• Intelligent Robots: State Estimation

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Accurate calibration of internal parameters is a crucial yet challenging prerequisite for 3D reconstruction using light field cameras. In this paper, we propose a linear fractional transformation(LFT) parameter $alpha$ to decoupled the main lens and micro lens array (MLA). The proposed method includes an analytical solution based on least squares, followed by nonlinear refinement. The method for detecting features from the raw images is also introduced. Experimental results on both physical and simulated data have verified the performance of proposed method. Based on proposed model, the simulation of raw light field images becomes faster, which is crucial for data-driven deep learning methods. The corresponding code can be obtained from the author's website.

Problem

• Calibrating internal parameters for light field cameras
• Decoupling main lens and micro lens array components
• Enabling faster simulation of raw light field images

Innovation

• Linear fractional transformation model decouples lens components
• Analytical least squares solution with nonlinear refinement
• Fast simulation method enables data-driven deep learning