Journal: Sensors (Basel, Switzerland)

Article Title: PIABC: Point Spread Function Interpolative Aberration Correction

doi: 10.3390/s25123773

Figure Lengend Snippet: Proposed end-to-end pipeline for sensor-based image restoration: The unboxed area illustrates the conventional image acquisition process, where unknown blur (PSF, point spread function) and sensor-induced electrical noise degrade image quality. Improving optical quality often leads to simultaneous enhancements in sensor performance, such as increased efficiency, reduced photon shot noise, and an extended dynamic range. In the proposed PIABC framework, spatially varying PSFs are first estimated from degraded sensor signals using a transformer-based interpolation method. These PSFs are then applied via Wiener filtering and further refined through residual correction to restore high-frequency details. The entire framework provides interpretable correction for optical aberrations and also demonstrates that, as a secondary benefit, optical restoration can reduce sensor-related artifacts. Here, η is sensor noise, and y * and h denote the patch of the degraded observation Y * = Y + η and the corresponding patch of the interpolated PSF field H respectively.

Article Snippet: Unlike post-processing approaches that rely on raytraced PSFs from Zemax and lens prescription data [ , ], PIABC takes a pre-processing perspective by simulating chromatic and spatially varying PSFs based on generic optical priors.

Techniques:

Journal: Sensors (Basel, Switzerland)

Article Title: PIABC: Point Spread Function Interpolative Aberration Correction

doi: 10.3390/s25123773

Figure Lengend Snippet: Concept of PIABC. ( a ) CS-sim PSF Patch (initial and interpolated): Each image patch is subdivided into four regions, from which CS-sim PSFs are embedded into a latent space using a trained autoencoder. A transformer decoder computes attention between a query position (marked with an arrow) and these sub-patches to interpolate the corresponding PSF. Query position t i is provided independently as input and serves as the query in the transformer cross-attention. PSF interpolation block: The attention output ( γ i j ) from the cross-attention is combined with position similarity weight ( α i j ) to produce interpolated PSF vectors ( z i j a , b , c , d ) , shown vertically stacked in the figure. ( b ) Image restoration, patch assembly, and deep correction: interpolated PSFs are used sequentially for Wiener deconvolution and residual correction to reconstruct the final restored image. All steps are described in detail in , and . For the overall query and correction structure, see ; for details of the transformer architecture, see , and for the residual correction pipeline, see . Note: Arrows are shown as examples, and only two are depicted here for clarity.

Article Snippet: Unlike post-processing approaches that rely on raytraced PSFs from Zemax and lens prescription data [ , ], PIABC takes a pre-processing perspective by simulating chromatic and spatially varying PSFs based on generic optical priors.

Techniques: Blocking Assay

Journal: Sensors (Basel, Switzerland)

Article Title: PIABC: Point Spread Function Interpolative Aberration Correction

doi: 10.3390/s25123773

Figure Lengend Snippet: Overview of the proposed PSF-aware transformer architecture: ( a ) Encoder: Each PSF patch is linearly embedded and combined with positional encodings (local and global) to form attention tokens. Attention is computed across sub-patches to capture spatial relationships. AE: Autoencoder. ( b ) Decoder: Queries at target positions are projected into the embedding space and attend to encoder features. In the decoder, the query at position i attends to encoder sub-patches at positions j , and the attention weights γ i j determine their relative contributions to the aggregated latent feature z i . Attention-weighted aggregation yields latent PSF representations, which are linearly transformed to generate spatially interpolated PSFs. Query configuration follows a. Sub-patch index ‘sp’ is omitted in for clarity but included here for detail. Layer Normalization (LN) is applied internally within the encoder. Decoder queries do not directly incorporate Local PE; instead, Global PE is indirectly propagated via attention; Case (a) in is used to illustrate the configuration. The bluish shaded blocks in the upper-right indicate subpatch-wise contributions to the interpolated result, based on the composite weight α i j · γ i j where α i j denotes position similarity and γ i j is the attention weight learned by the transformer; Relevant positional encodings and notations are indicated by shade intensity in the figure and fully defined in .

Article Snippet: Unlike post-processing approaches that rely on raytraced PSFs from Zemax and lens prescription data [ , ], PIABC takes a pre-processing perspective by simulating chromatic and spatially varying PSFs based on generic optical priors.

Techniques: Transformation Assay

Journal: Sensors (Basel, Switzerland)

Article Title: PIABC: Point Spread Function Interpolative Aberration Correction

doi: 10.3390/s25123773

Figure Lengend Snippet: Patch assembly and compilation pipeline using Wiener Filter image restoration: The left part illustrates the input aberrated image patches and the interpolated PSFs generated by the Transformer for each query configuration (see ). The center part shows how the interpolated PSFs are combined with the Wiener filter to reconstruct image patches under each setting. The right part depicts α-weighted fusion of reconstructed results from CS-sim and interpolated PSFs, followed by patch integration, deep feature correction using a residual U-Net (see ), and final full-size image restoration.

Article Snippet: Unlike post-processing approaches that rely on raytraced PSFs from Zemax and lens prescription data [ , ], PIABC takes a pre-processing perspective by simulating chromatic and spatially varying PSFs based on generic optical priors.

Techniques: Generated

Journal: Sensors (Basel, Switzerland)

Article Title: PIABC: Point Spread Function Interpolative Aberration Correction

doi: 10.3390/s25123773

Figure Lengend Snippet: Stage-wise qualitative comparison of restored images: Each row shows a test image from the DIV2K dataset processed through four stages of our restoration pipeline: ( a ) Ground Truth; ( b ) Aberrated image; ( c ) Final output after residual U-Net correction; ( d ) Wiener-filtered image using observed PSFs; ( e ) Patch-wise compilation using interpolated PSFs. Below each row, red-boxed regions are cropped and enlarged for detailed comparison. For ( d , e ), the applied PSFs are also visualized on the right. Interpolated PSFs appear smoother and more spatially coherent due to softmax-weighted blending and softplus regularization, which helps reduce aliasing at the cost of some high-frequency sharpness. This facilitates more effective deep correction using residual U-Net.

Article Snippet: Unlike post-processing approaches that rely on raytraced PSFs from Zemax and lens prescription data [ , ], PIABC takes a pre-processing perspective by simulating chromatic and spatially varying PSFs based on generic optical priors.

Techniques: Comparison