Samsung Electronics proposes motion correction for photon-counting image sensors. US application US20260313265A1, published on October 8, 2026, aligns short subframes with a reference image before combining them into a frame. This aims to reduce blurring caused by adding different positions of moving subjects.
Align first, then add
In a SPAD sensor, a detected photon triggers an electrical avalanche whose signal is recorded by an in-pixel counter. The proposed circuit creates several subframes within a frame from counts accumulated over short exposure units. A processor selects a reference subframe and calculates per-pixel displacement vectors for other subframes. It corrects their positions and adds the aligned subframes to the reference image. Stronger motion can increase the subsequent subframe rate; limited motion can reduce it. Different rates inside and outside selected image regions are also proposed. This aims to limit unnecessary processing.
The original document describes alignment and addition in independent claims 1 and 10, paragraphs [0102]–[0109] and Figure 14. Claim 17 adds the motion-dependent subframe rate.
Dark pixels remain a limitation
Displacement vectors in dark regions can be unreliable because contrast is insufficient. Paragraphs [0174]–[0176] and claim 7 provide for excluding such pixels from vector calculation. The document supplies no reproducible benchmark for sharpness improvement or energy savings and confirms no production sensor. Example exposure times and subframe counts are not measured product performance. Quality under real capture conditions remains unverified.
Document and status
“Image Sensor Including Avalanche Photodiode”: Applicant Samsung Electronics Co., Ltd.; US application 19/638,514, filed April 3, 2026; claimed Japanese priority 2025-062894 of April 7, 2025. A1 denotes a published application, not a grant. We could not independently confirm the current official prosecution status on October 10, 2026.
Pandorex View: Combining photon counting with subframe alignment could interest cameras capturing moving subjects. The key question is whether reliable motion information comes at a reasonable computational cost. The new publication establishes the proposed mechanism; its performance requires practical testing.