Defocus Aberration Theory in Imaging

• Defocus Aberration Theory is defined as the loss of image sharpness caused by a quadratic phase error in the lens's pupil function.
• The theory uses Gaussian and Fourier-based models to accurately describe the point spread function and predict spatial blurring across different imaging devices.
• Data-driven methods, including neural networks and self-supervised approaches, enhance the estimation and correction of defocus aberrations for improved imaging performance.

Defocus aberration theory addresses the fundamental optical and mathematical description of image degradation—specifically, the loss of sharpness resulting when an imaging system forms an image of a point object that does not lie in the lens’s focal plane. The central concept is that defocus introduces a phase error that manifests as a quadratic phase aberration in the system’s pupil function, leading to spatial blurring in the image described analytically and computationally via the point spread function (PSF) and optical transfer function (OTF). Defocus plays a critical role in various imaging modalities, including conventional photography, microscopy, optical coherence tomography (OCT), and x-ray spectroscopy, with direct implications for computational refocusing, depth estimation, and aberration correction.

1. Mathematical Framework: Defocus in the Pupil Function and Image Formation

The optical system's pupil function with defocus is modeled as

$$P(\rho) = A(\rho)\exp\left[i\frac{2\pi}{\lambda}k z \rho^2\right]$$

where $\rho$ is the normalized pupil coordinate, $A(\rho)$ is the amplitude (e.g., circular or Gaussian aperture), $z$ is the defocus distance (object-plane offset from focus), $\lambda$ is the wavelength, and $k$ is a lens-specific scaling factor. This quadratic phase term explicitly encodes defocus aberration as a second-order Seidel/Zernike term (Zhu et al., 23 Jan 2025, Makita et al., 25 Jan 2025).

The resultant complex amplitude PSF, the Fourier transform of $P(\rho)$, quantifies the lateral response: $$h(x;z) = \int A(\rho)\exp\left[i\frac{2\pi}{\lambda}k z \rho^2\right] \exp\left[-i\frac{2\pi}{\lambda f}\rho x\right] d\rho$$ For a Gaussian pupil, this yields a Gaussian PSF whose beam radius broadens with defocus as $w(z)=w_0\sqrt{1+(z/z_R)^2}$. The intensity PSF accordingly expands as

$$|h(x;z)|^2 \propto \exp\left[-2\frac{x^2}{w^2(z)}\right]$$

This broadening encapsulates the lateral image blur induced by defocus.

In spatial-frequency (Fourier) coordinates, the coherent transfer function (CTF) in the presence of defocus is expressed as

$H(f;z) = A(f) \exp\left[-i\pi\lambda z f^2\right]$

where the quadratic phase factor is the classical defocus filter, enabling analytic phase correction in computational refocusing (Zhu et al., 23 Jan 2025, Makita et al., 25 Jan 2025).

2. Point Spread Function Modeling and Gaussian Approximation

The PSF for defocus aberration in practical, diffraction-limited systems is accurately approximated by a spatially-invariant Gaussian kernel: $$h(x,y) = \frac{1}{2\pi \sigma^2} \exp\left(-\frac{x^2 + y^2}{2\sigma^2}\right)$$ with $\sigma$ proportional to the circle of confusion diameter $C$. For most modern imaging devices, the empirically-validated Gaussian model provides mean absolute error (MAE) $<1\%$ for typical focus depths of $1$–$100$ m and defocus up to $10\%$ of focused depth, capturing defocus transfer functions with high fidelity (Saadat, 8 Jan 2026, He et al., 2022). This single-parameter Gaussian blur captures both the absolute and the relative blur between two differently focused images, establishing a well-posed analytic framework for depth-from-defocus (DFD).

A table summarizing the accuracy of the Gaussian PSF model:

Device Type	Focus Range (m)	Max Defocus (%)	MAE (Gaussian Fit)
Conventional Cameras	1–100	≤10	< 1%
Pixel Pitches ≤ 5.6μm	1–100	≤10	< 1%

3. Spatial Variation, Aberrations, and Data-Driven Estimation

Defocus aberration theory extends to non-ideal lenses by explicitly modeling spatially-varying, rotationally symmetric PSFs parameterized by image coordinates, depth, and focus settings: $p(\Delta x,\Delta y;\,x,y,d,f_d)$ This PSF is often estimated in a polar basis to leverage symmetry in rotationally symmetric systems and is learned using self-supervised or non-blind approaches with data from sharp and blurred images (Wu et al., 2024, Lin et al., 2023). Neural representations such as multilayer perceptrons (MLP) model variations of the PSF as a continuous high-dimensional function of spatial coordinates, depth, focus, and sub-pixel offsets (Lin et al., 2023). Such lens blur fields capture higher-order aberrations beyond pure defocus, such as coma, astigmatism, and manufacturing variabilities.

These models support artifact-free reconstruction, realistic simulation of bokeh, and device fingerprinting for optical characterization.

4. Computational Refocusing and Limits in OCT and Microscopy

In computational imaging, defocus can be corrected by applying a phase-conjugate filter in the Fourier domain: $$\text{Refocusing filter:} \quad \exp\left[+i\pi\lambda z f^2\right]$$ This removes the defocus-induced quadratic phase and retrieves a sharp, in-focus image (Zhu et al., 23 Jan 2025, Makita et al., 25 Jan 2025).

For optical coherence tomography (OCT), the limitations of computational refocusing are theory-backed and modality-dependent:

• In point-scanning (confocal) OCT, the maximum correctable defocus (MCD) is limited by confocality (signal attenuation due to the off-focus Gaussian envelope), scaling as $MCD_{scan} = b^2/(2n\lambda)$, where $b$ is the confocal parameter (Zhu et al., 23 Jan 2025).
• Spatially coherent full-field OCT (FFOCT) lacks the confocal gate and presents no such limit (i.e., infinite MCD apart from sampling constraints), enabling computational refocusing over arbitrarily large defocus ranges (Zhu et al., 23 Jan 2025, Makita et al., 25 Jan 2025).

In microscopy, digitally induced interference fringes in summed Fourier spectra from two-angle illumination (DAbI) directly encode defocus through a physics-based relation, allowing quantitative defocus measurement and effective depth-of-field extension by factors up to 20 (Zhou et al., 15 Jul 2025).

5. Experimental Estimation and Optimization of Defocus PSF

Joint estimation of the defocus PSF parameters employs loss functions constructed from luminance, Laplacian (sharpness), defocus-histogram, and structural similarity (SSIM) terms: $$L = \alpha_1 L_1 + \alpha_2 L_2 + \alpha_3 L_3 + \alpha_4 L_4$$ with empirical weighting. The model typically assumes a spatially varying, Gaussian-shaped PSF, with variance $r(x,y)$ parameterized as

$$r(x,y) = A \frac{|D_{gt}(x,y) - D_f|}{D_{gt}(x,y)(D_f - F)}$$

where $A$ embeds all optical and calibration constants. Optimal parameters $(A^*, e^*)$ are found via exhaustive search or gradient-based minimization over focused image stacks with known ground-truth depth and all-in-focus reference (He et al., 2022).

Recent approaches train deep networks to predict field- and depth-dependent PSFs directly from data, incorporating spatial prior constraints (radial monotonicity, smoothness) and handling focus breathing effects during focal stack acquisition (Wu et al., 2024, Lin et al., 2023).

6. Practical and Theoretical Implications in X-ray and Optical Imaging

Defocus aberration manifests as an energy-dependent quadratic phase error in x-ray echo spectrometers. For lens-based focusing elements, the defocus term vanishes ($\tan\theta \to 0$), yielding truly aberration-free imaging. Mirror-based systems exhibit a linear variation in best focus with energy transfer; this is corrected by tilting the detector plane by an angle that compensates the energy-dependent focus shift, restoring diffraction-limited resolution (Río et al., 2018).

These theoretical results prescribe the system design parameters—numerical aperture (NA), focusing geometry, and mirror slope tolerances—to ensure minimal defocus-induced resolution loss.

7. Theoretical Assumptions and Applicability Limits

Defocus aberration theory relies on several critical assumptions:

• Paraxial (small-angle) approximation; high-NA corrections and non-paraxial effects are not included.
• Rotational symmetry of the lens and PSF; off-axis and asymmetric aberrations require additional modeling.
• Diffraction-limited, narrow-band, and spatially coherent illumination, particularly relevant for analytic Gaussian PSF models and computational refocusing.
• Purely geometric optics or scalar wave optics, neglecting polarization and multiple scattering.
• Negligible high-order aberrations for standard models; neural and data-driven methods capture these if present (Zhu et al., 23 Jan 2025, Saadat, 8 Jan 2026, Lin et al., 2023).

Violations, such as high-NA systems, strong field curvature, vignetting, or sensor nonuniformities, require more comprehensive aberration modeling (e.g., full Zernike expansion, vectorial Debye integral).

---

References: (Zhu et al., 23 Jan 2025, Makita et al., 25 Jan 2025, Wu et al., 2024, Lin et al., 2023, Saadat, 8 Jan 2026, Zhou et al., 15 Jul 2025, He et al., 2022, Río et al., 2018)