Generic Camera Calibration using Blurry Images

Zezhun Shi

cs.CVMar 5, 2026

Generic Camera Calibration using Blurry Images

Authors: Zezhun Shi

Organizations: Independent Researcher

Abstract

Camera calibration is the foundation of 3D vision. Generic camera calibration can yield more accurate results than parametric cam era calibration. However, calibrating a generic camera model using printed calibration boards requires far more images than parametric calibration, making motion blur practically unavoidable for individual users. As a f irst attempt to address this problem, we draw on geometric constraints and a local parametric illumination model to simultaneously estimate feature locations and spatially varying point spread functions, while re solving the translational ambiguity that need not be considered in con ventional image deblurring tasks. Experimental results validate the effectiveness of our approach.

Figures & tables

Figure 1: Parametric model and generic model. The latter can yield more accurate results than former but requires far more calibration images.

Figure 2: Generic camera calibration requires far more images than parametric calibration, making motion blur practically unavoidable. Standard deconvolution recovers visually plausible images but leaves feature positions ambiguous up to an arbitrary translation due to the shift equivariance of convolution.

Figure 3: Overview of the proposed method. We use local homographies and illumination parameters to replace classical feature based global mapping to describe the latent image, with geometric constraints enforcing inter-block consistency. Global translational ambiguity is resolved by alignment with a calibrated parametric camera.

Figure 4: Deconvolution results for four PSF glyphs under checkerboard and star calibration patterns, with and without 5% Gaussian noise. Each group shows an $80\times 80$ and $160\times 160$ px crop with SSIM/PSNR. The star pattern is significantly more robust to noise.

Figure 5: Distribution of reprojection direction (hue) and magnitude.

Figure 6: Transition bias distribution against median reprojection error of raw data (per frame).

Angle Threshold	0			$\pi$ /10
Loss Type	Huber	Mean	Median	Huber	Mean	Median
Random Shifted	0.257/0.275	0.259/0.278	0.260/0.278	0.236/0.279	0.243/0.286	0.243/0.286
Global Aligned	0.061 / 0.064	0.064/0.068	0.090/0.096	0.042 / 0.044	0.058/0.059	0.065/0.066
Random Shifted(self)	0.213/0.278	0.213/0.278	0.213/0.278	0.289/0.354	0.288/0.354	0.288/0.354
Global Aligned(self)	0.036 / 0.050	0.037/0.051	0.039/0.057	0.036 / 0.047	0.037/0.048	0.039/0.052
Raw	0.028/0.036	0.028/0.036	0.028/0.036	0.031/0.040	0.031/0.039	0.031/0.039

Table 1: Medium/mean reprojection errors comparison (px).

Figure 7: Estimated spatially varying PSF field for a sample blurry image. Left: without centroid correction. Right: with corrective recentering, showing spatially coherent motion blur that varies smoothly across the field. Bottom row: magnified PSFs from the red boxed region.

Figure 8: Cross image PSF quality analysis. Left: loss vs. blur size, colored by eccentricity. Right: loss vs. boundary energy ratio, colored by $\sigma_{\mathrm{major}}$ . High boundary energy indicates PSF truncation and correlates with elevated loss, motivating the element level filtering.

Figure 9: Median reprojection error after global alignment with bilinear bias field (Eq. 13 ). Error bars denote interquartile range. Local alignment with BE filtering (red) achieves

{\sim}\,0.08

px at

\tau_{L}{=}30

, outperforming both raw translation (gray) and BE only filtering (blue).

Camera Calibration

Camera calibration estimates the geometric and optical parameters that connect image measurements to real-world positions and other sensors.

See all 57 papers

UniDeblur: Unified Motion Deblurring via Fourier Convolution Learning
Oct 9, 2026Xintian Mao, Jiansheng Wang, Huaiyuan Wang +2Image Deblurring Frequency-Domain Image Processing
Shanghai Key Laboratory of Multidimensional Information Processing, East China Normal University, China. · Shanghai Institute of Technical Physics, Chinese Academy of Sciences, China.
Controllable blind deblurring with diffusion models
Aug 24, 2026Imane Si Salah, Emile Cribelier, Thomas Veit +2Image Deblurring Image Restoration
DxO Labs, France · LTCI, Télécom Paris, IP Paris, France
Unblur-SLAM: Dense Neural SLAM for Blurry Inputs
Mar 26, 2026Qi Zhang, Denis Rozumny, Francesco Girlanda +4Visual SLAM Camera Pose Estimation
University of Amsterdam, Amsterdam, Netherlands · Meta Reality Labs, Zurich, Switzerland · ETH Zürich, Zürich, Switzerland

Oct 9, 2026J/K move · Enter open · S save

UniDeblur: Unified Motion Deblurring via Fourier Convolution Learning

Xintian Mao, Jiansheng Wang, Huaiyuan Wang, Qingli Li+1

Shanghai Key Laboratory of Multidimensional Information Processing, East China Normal University, China. · Shanghai Institute of Technical Physics, Chinese Academy of Sciences, China.

Image Deblurring Frequency-Domain Image Processing

Open paper AskPDF arXiv

Image deblurring is a fundamental task in computer vision, yet existing methods mainly specialize in isolated scenarios. Consequently, the deblurring performance of these models drops significantly across scenarios with different cameras and image contents. This motivates the development of a unified motion deblurring framework that can handle diverse blur patterns. To this end, we construct a multi-source dataset and develop a data pruning and quality control pipeline. Then, to restore blur images with various motion blur kernels across different scenarios, we propose a unified motion deblurring framework via Fourier Convolution Learning. Existing deblurring networks usually rely on pixel-level mappings and overlook the convolutional nature of blur. Thus, the network mostly fits the image content itself rather than learning the convolution process, making it difficult to handle multi-source blur images. Instead of relying on spatial mapping, we perform blur-adaptive global filter learning in Fourier space, which avoids the limited spatial support of kernel estimation. Extensive experiments demonstrate that our unified model effectively handles motion blurs across diverse sources and scenes, achieving competitive performance across both in-domain and OOD benchmarks. Code has been released at github.com/DeepMed-Lab-ECNU/Single-Image-Deblur.

Abstract

Camera Calibration

Explore similar work

UniDeblur: Unified Motion Deblurring via Fourier Convolution Learning

Controllable blind deblurring with diffusion models

Unblur-SLAM: Dense Neural SLAM for Blurry Inputs