Speckle-Learned Recognition (SLR)

• SLR is a technique that uses machine learning to extract information from speckle fields generated by scattering media and structured light.
• It leverages diverse learning models—from supervised CNNs to metric and unsupervised methods—to analyze 2D, 1D spatial, and temporal speckle data even under noisy or misaligned conditions.
• SLR finds applications in areas such as optical demultiplexing, non-line-of-sight recognition, biometric encryption, and material identification, achieving high accuracy with minimal data.

Searching arXiv for the provided SLR-related papers and closely related work to ground the article. arxiv_search(query="^^^^^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^^^^^", max_results=^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, sort_by="^^^^^^^^7relevance7 arxiv_search(query="^^^^^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^^^^^", max_results=^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, sort_by="^^^^^^^^7relevance7 arxiv_search(query="^^^^^^^^7\7 ^^^^^^^^7\7 imaging^^^^^^^^7\7 OR ^^^^^^^^7\7 learning meets Singular Optics: Speckle-based Structured light demultiplexing^^^^^^^^7\7 max_results=^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, sort_by="^^^^^^^^7relevance7 Speckle-Learned Recognition (SLR) denotes a class of inference methods in which machine learning models identify physical states, objects, or communication symbols from speckle fields rather than from directly formed images. In these systems, speckles generated by scattering media, diffusers, walls, random phase masks, or detector-limited diffraction patterns are treated as information-bearing fingerprints, and the learning stage can be supervised, metric-learning-based, or unsupervised. Reported instantiations span structured-light demultiplexing, X-ray single-particle imaging (SPI), non-line-of-sight object recognition, optical communication, biometric encryption, material recognition, and pose tracking, with ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D spatial, ^^^^7relevance7^^^^D spatial, and ^^^^7relevance7^^^^D temporal measurements all used as recognition substrates (&&&^^^^7relevance7^^^^&&&, &&&^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&, &&&^^^^7relevance7^^^^&&&, Badavath et al., 21 Sep 2025).

^^^^7relevance7^^^^. Physical basis and signal formation

SLR rests on the observation that a speckle field is not merely random noise. In structured-light settings, speckle fields are generated by the interference of multiple wavefronts diffracted and scattered through a diffuser, producing a random distribution whose phase and intensity correlate to the incident beam. A canonical model writes the far-field speckle as

PRESERVED_PLACEHOLDER_^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^

where PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^ is the input structured-light field, PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ is the random phase imparted by the diffuser, and PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^ is the ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D Fourier transform (&&&^^^^7relevance7^^^^&&&).

In orbital angular momentum (OAM) demultiplexing, the same principle is extended by deliberately modifying the optical transformation. The beam first passes through a random phase mask and then a tilted spherical convex lens that introduces astigmatism. The astigmatic lens transformation is

PRESERVED_PLACEHOLDER_^^^^7\7^^^^

with PRESERVED_PLACEHOLDER_^^^^7 \7^^^^ and PRESERVED_PLACEHOLDER_^^^^7 OR \7^^^^, and the far-field intensity is $I_{sf}(x,y;f)=|U_{sf}(x,y;f)|^2$. A tilt of $20^\circ$–$40^\circ$, used at PRESERVED_PLACEHOLDER_^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, breaks the intensity degeneracy of conjugate pairs such as PRESERVED_PLACEHOLDER_^^^^7relevance7relevance7^^^^ and PRESERVED_PLACEHOLDER_^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^, making their speckle patterns distinguishable (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&).

In non-line-of-sight (NLOS) recognition, coherent illumination scattered by hidden objects and relayed by a diffusive wall likewise produces object-dependent speckles. The recorded intensity at a point PRESERVED_PLACEHOLDER_^^^^7relevance7relevance7^^^^ is modeled as

PRESERVED_PLACEHOLDER_^^^^7relevance7\7^^^^

where PRESERVED_PLACEHOLDER_^^^^7relevance7 \7^^^^ is the field amplitude from object point PRESERVED_PLACEHOLDER_^^^^7relevance7 OR \7^^^^ to image point PRESERVED_PLACEHOLDER_^^^^7relevance77^^^^, and PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^8 encodes geometry-dependent phase information (&&&^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&).

Later SLR work makes the representational point explicit by distinguishing a ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D spatial speckle field PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^9, a ^^^^7relevance7^^^^D spatial speckle array PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ or PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7relevance7^^^^, and a ^^^^7relevance7^^^^D temporal speckle sequence PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^. This shows that ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D spatial information of structured light can be mapped onto lower-dimension speckle data in space or time (Badavath et al., 21 Sep 2025).

^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^. Learning formulations

A large part of SLR uses supervised discriminative models. For structured-light demultiplexing, AlexNet is used with ^^^^7 \7^^^^ convolutional feature-extraction stages and a final softmax classifier; for OAM demultiplexing the last layer is modified for ^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^-class classification, and the network is fine-tuned with Stochastic Gradient Descent with Momentum at learning rate PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7relevance7^^^^ on ^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ simulated speckle images per class, split 8^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^%/^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^% for training and validation/testing (&&&^^^^7relevance7^^^^&&&, &&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&). In NLOS recognition, SimpleNet is used for handwritten digits and ResNet-^^^^7relevance7^^^^8 for human body posture classification, both trained directly on speckle images rather than reconstructed scenes (&&&^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&).

A distinct formulation is metric learning. SpeckleNN converts SPI speckle classification from direct multi-class prediction to learning a unified embedding vector space in which similarity is measured by Euclidean distance. Its embedding model uses ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ convolutional layers followed by ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ fully connected layers and outputs a ^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8-dimensional embedding constrained to unit length. Training uses triplets PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7\7^^^^ and the triplet loss

PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 \7^^^^

with semi-hard triplet mining, Adam optimizer, learning rate PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 OR \7^^^^, and data augmentation by random in-plane rotation, masking, zooming, and shifting. Few-shot classification then assigns a query PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)77^^^^ by

PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8

using the class with minimum average support distance. The paper emphasizes that new classes or samples can be introduced without retraining, provided a few labeled examples are available, and that the method scales linearly with dataset size (&&&^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&).

SLR is not restricted to supervised training. The unsupervised framework "Speckle Unsupervised Recognition and Evaluation" (SURE) introduces two routes. SCAN, used for multiple or thick scattering, learns invariant features with an Invariant Information Clustering loss,

PRESERVED_PLACEHOLDER_^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^9

while SHACK, used for single or thin scattering, performs hierarchical agglomerative clustering with Euclidean distance on vectorized speckle images. In this formulation, semantic labels can be assigned after clustering by labeling a representative example per cluster (&&&^^^^7relevance7^^^^&&&).

At the resource-constrained end, SLR also includes domain-specific compact classifiers. A lightweight CNN for speckle-based material recognition uses only the green channel at PRESERVED_PLACEHOLDER_^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ resolution, a first PRESERVED_PLACEHOLDER_^^^^7relevance7relevance7^^^^ convolution with ^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ filters and stride ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^, depthwise convolution, pointwise PRESERVED_PLACEHOLDER_^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ convolutions with ^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8 and ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 \7 OR \7^^^^ filters, global average pooling, and fully connected layers of ^^^^7 \7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^, ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 \7 OR \7^^^^, ^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8, and ^^^^7 \7^^^^9 neurons. The model contains ^^^^7relevance7\7relevance7^^^^,^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN77^^^^ trainable parameters, approximately PRESERVED_PLACEHOLDER_^^^^7relevance7relevance7^^^^ MB (&&&^^^^7relevance77^^^^&&&).

^^^^7relevance7^^^^. Data efficiency, robustness, and reduced-order sensing

A central result in SLR is that high recognition accuracy does not necessarily require large labeled datasets. SpeckleNN is explicitly designed for limited labeled examples and reports effective few-shot behavior: with as few as ^^^^7 \7^^^^ labeled examples per class, performance is similar to using ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ or more per class. The same system is reported to remain effective even when up to 7^^^^7 \7^^^^% of the detector area is missing or masked, with 9^^^^7\7^^^^% accuracy and ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.9^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ F-^^^^7relevance7^^^^ score using only ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 \7^^^^% area, compared with 7^^^^7\7^^^^% accuracy for the prior model (&&&^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&).

Structured-light SLR emphasizes distributed information content. Recognition is reported from a small portion of the speckle field, and this partial-field property supports off-axis operation. In one study, on-axis simulated accuracy exceeds 99% for both LG and HG modes, experimental accuracy is 9^^^^7 OR \7^^^^%, accuracy for degenerate LG modes rises from approximately ^^^^7 \7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^% without astigmatic transformation to approximately 99% with a cylindrical lens, off-axis recognition gives 9^^^^7 OR \7^^^^% on-axis and 9^^^^7relevance7^^^^% at PRESERVED_PLACEHOLDER_^^^^7relevance7\7^^^^ off-axis, and accuracy remains about 98% under Gaussian white noise at SNR PRESERVED_PLACEHOLDER_^^^^7relevance7 \7^^^^ dB and under simulated atmospheric turbulence with PRESERVED_PLACEHOLDER_^^^^7relevance7 OR \7^^^^ (&&&^^^^7relevance7^^^^&&&).

The astigmatic OAM demultiplexing line makes a related claim in communication settings: speckle-learned recognition is alignment-insensitive, described as alignment-free to first order, and more resilient to misalignment, turbulence, and noise than direct-intensity-based classification because it relies on global speckle statistics rather than perfect capture of the original mode profile (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&).

Reduced-order sensing extends these ideas to ^^^^7relevance7^^^^D and single-pixel acquisition. For ^^^^7relevance7^^^^D spatial SLR, ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ line profiles of length PRESERVED_PLACEHOLDER_^^^^7relevance77^^^^ pixels are acquired per class and classified by a custom ^^^^7relevance7^^^^D CNN; peak accuracy reaches 98% for LG modes, and accuracy remains above 9^^^^7\7^^^^% even when using PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^8th of the ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D speckle data. For ^^^^7relevance7^^^^D temporal SLR, a support vector machine classifies temporal speckle sequences collected by a single-pixel detector or virtual detector grid; reported accuracies are up to 9^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^.^^^^7 \7^^^^% for HG and 8^^^^7 OR \7^^^^.7% for LG with a PRESERVED_PLACEHOLDER_^^^^7relevance7^^^^9 pixel detector, and exceed 9^^^^7 OR \7^^^^% across structured-light families with optimized conditions. Recognition improves when PRESERVED_PLACEHOLDER_^^^^7\7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, and under strong turbulence modeled with PRESERVED_PLACEHOLDER_^^^^7\7relevance7^^^^, temporal SLR remains above 9^^^^7 \7^^^^% when PRESERVED_PLACEHOLDER_^^^^7\7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ (Badavath et al., 21 Sep 2025).

^^^^7\7^^^^. Major application domains

Reported SLR systems span substantially different sensing geometries and objectives. The following results are therefore task-specific rather than directly comparable.

Setting	Measurement / model	Reported result
XFEL SPI speckle classification	^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8-dimensional unified embedding with triplet learning	9^^^^7\7^^^^% accuracy and ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.9^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ F-^^^^7relevance7^^^^ score with only ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7 \7^^^^% detector area; effective ^^^^7 \7^^^^-shot classification (&&&^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&)
Structured-light demultiplexing	^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D speckle images with AlexNet	>99% simulated accuracy, 9^^^^7 OR \7^^^^% experimental accuracy, and 9^^^^7relevance7^^^^% at PRESERVED_PLACEHOLDER_^^^^7\7relevance7^^^^ off-axis (&&&^^^^7relevance7^^^^&&&)
Astigmatic OAM shift keying and multiplexing	^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^-class AlexNet on far-field astigmatic speckles	>98% accuracy for both OAM shift keying and OAM multiplexing in simulated results (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&)
^^^^7relevance7^^^^D spatial and temporal structured-light SLR	^^^^7relevance7^^^^D CNN on spatial arrays; SVM on temporal sequences	98% peak LG accuracy for ^^^^7relevance7^^^^D spatial SLR, >9^^^^7\7^^^^% with PRESERVED_PLACEHOLDER_^^^^7\7\7^^^^th of ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D data, and >9^^^^7 OR \7^^^^% temporal accuracy with optimized conditions (Badavath et al., 21 Sep 2025)
NLOS object recognition	Speckle images on visible walls with SimpleNet or ResNet-^^^^7relevance7^^^^8	9^^^^7 \7^^^^% one-wall experimental accuracy, 9^^^^7relevance7^^^^% with rotating wall, and 78.^^^^7relevance7^^^^8% mean accuracy for human body posture simulation (&&&^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&)
Label-free speckle clustering and decoding	SCAN or SHACK without labeled training data	9^^^^7 \7^^^^% clustering accuracy for time-lapse glucose concentrations and 97% accuracy for a ^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^-class multimode-fiber communication task (&&&^^^^7relevance7^^^^&&&)
Speckle-based face recognition under encryption	Optical speckle encryption plus U-Net decryption and ResNet-based face encoding	>98% face recognition accuracy, specifically 98.^^^^7\7^^^^9% at threshold ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.^^^^7 \7^^^^8 (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8&&&)
Speckle-based material recognition for laser cutting	Lightweight ^^^^7relevance7\7relevance7^^^^k-parameter CNN on SensiCut	9^^^^7 \7^^^^.^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7 \7^^^^% test accuracy, macro and weighted F^^^^7relevance7^^^^-scores of ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.9^^^^7 \7relevance7^^^^, and ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7Speckle Unsupervised Recognition and Evaluation7 \7^^^^ images per second (&&&^^^^7relevance77^^^^&&&)

This suggests that SLR is best understood as a reusable inference strategy across classification, demultiplexing, clustering, identification, and, in adjacent work, regression, rather than as a single optical architecture.

^^^^7 \7^^^^. System-level extensions

SLR has also been integrated into privacy-preserving and system-level optical pipelines. In a speckle-based optical cryptosystem for face recognition, a scattering ground glass acts as a physical secret key and the forward model is written as

PRESERVED_PLACEHOLDER_^^^^7\7 \7^^^^

where PRESERVED_PLACEHOLDER_^^^^7\7 OR \7^^^^ is the vectorized face image, PRESERVED_PLACEHOLDER_^^^^7\77^^^^ is the transmission matrix of the scattering system, and PRESERVED_PLACEHOLDER_^^^^7\7^^^^8 is the measured speckle pattern. Decryption is performed by a U-Net with an additional complex fully connected layer and a normalization layer, trained on ^^^^7relevance7^^^^9,8^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ paired speckle-face images using a loss that combines mean squared error and negative Pearson correlation coefficient. The reported key length is PRESERVED_PLACEHOLDER_^^^^7\7^^^^9, approximately ^^^^7relevance77^^^^.^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ Gbits, decryption remains effective from a quarter field of view and under moderate Gaussian noise up to ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.^^^^7relevance7^^^^ standard deviation, and end-to-end face recognition exceeds 98% accuracy (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^8&&&).

A second extension moves from recognition to tracking. SpecTrack uses Laser Speckle Imaging with a lensless camera and a retro-reflector marker carrying a coded aperture, then learns a mapping from FFT-transformed speckle frames to PRESERVED_PLACEHOLDER_^^^^7 \7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^, PRESERVED_PLACEHOLDER_^^^^7 \7relevance7^^^^, and PRESERVED_PLACEHOLDER_^^^^7 \7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^. The network takes ^^^^7 \7^^^^ consecutive FFT-transformed frames concatenated into a tensor of shape PRESERVED_PLACEHOLDER_^^^^7 \7relevance7^^^^, applies three convolutional blocks and a six-layer multilayer perceptron, and reports accuracy of PRESERVED_PLACEHOLDER_^^^^7 \7\7^^^^ with standard deviation PRESERVED_PLACEHOLDER_^^^^7 \7 \7^^^^, y-axis mean absolute error PRESERVED_PLACEHOLDER_^^^^7 \7 OR \7^^^^ with standard deviation PRESERVED_PLACEHOLDER_^^^^7 \77^^^^, z-axis mean absolute error PRESERVED_PLACEHOLDER_^^^^7 \7^^^^8 with standard deviation PRESERVED_PLACEHOLDER_^^^^7 \7^^^^9, depth accuracy PRESERVED_PLACEHOLDER_^^^^7 OR \7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ cm, and frame rate ^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^ FPS (&&&^^^^7relevance7relevance7^^^^&&&).

An adjacent strand concerns learning the speckle patterns themselves rather than only learning to decode them. Speckle-Net is a multi-branch, two-layer convolutional framework for speckle pattern design in computational ghost imaging, optimized by backpropagating imaging loss to the speckle-generation kernels. It reports high-quality reconstructions with sampling ratio PRESERVED_PLACEHOLDER_^^^^7 OR \7relevance7^^^^ as low as ^^^^7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^.^^^^7 \7^^^^%, and the paper states that altering the loss function could optimize the same framework for recognition targets rather than image reconstruction (&&&^^^^7relevance7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&).

^^^^7 OR \7^^^^. Misconceptions, limitations, and research directions

A common misconception is that speckles are too random to support high-level inference. Across the reported literature, the opposite claim is repeatedly demonstrated: speckle fields behave as unique statistical fingerprints of structured light, materials, hidden objects, or encrypted biometrics, provided that the optical transformation and the learning model are matched to the task (&&&^^^^7relevance7^^^^&&&). A second misconception is that recognition requires full ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D speckle capture and precise alignment. Reported systems operate from a small portion of the speckle field, at off-axis positions, from ^^^^7relevance7^^^^D spatial arrays, and from ^^^^7relevance7^^^^D temporal sequences recorded by a single-pixel detector (Badavath et al., 21 Sep 2025).

Another misconception is that SLR is inherently supervised. The SURE framework shows that clustering and evaluation can proceed without labeled datasets by extracting invariant features from speckles and assigning labels only after clustering. This does not remove the need for validation, but it does show that label-free recognition and decoding are viable in dynamic scattering environments (&&&^^^^7relevance7^^^^&&&).

Several limitations also recur. In NLOS recognition, performance depends strongly on the architecture and training of the deep network; physical movement, vibration, or drift can shift speckle patterns; signal intensity drops after multiple diffusive reflections; and recognition of unseen object types not covered in training is explicitly described as unexplored (&&&^^^^7relevance7Speckle-Learned Recognition speckle structured light demultiplexing SpeckleNN7^^^^&&&). In structured-light systems, intensity-degenerate modes such as PRESERVED_PLACEHOLDER_^^^^7 OR \7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^ and PRESERVED_PLACEHOLDER_^^^^7 OR \7relevance7^^^^ require an additional astigmatic transformation to become separable in practice (&&&^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^&&&). Even where robustness is strong, it is not uniform across sensing modalities: spatial SLR is reported to degrade to approximately 79% at PRESERVED_PLACEHOLDER_^^^^7 OR \7\7^^^^, whereas temporal SLR remains above 9^^^^7 \7^^^^% in strong turbulence when PRESERVED_PLACEHOLDER_^^^^7 OR \7 \7^^^^ (Badavath et al., 21 Sep 2025).

A plausible implication is that robust SLR is fundamentally a co-design problem. The reported successes combine optical transformations that preserve or expose discriminative invariants, detector geometries that retain useful speckle statistics, and learning objectives matched to the inference task—classification, metric retrieval, clustering, decryption, or regression. Within that framing, the field has already moved from direct ^^^^7(Wang et al., 2023) OR (Raskatla et al., 2023) OR (Das et al., 2024) OR (Fan et al., 2024)7^^^^D CNN classification to few-shot embedding models, label-free clustering, single-pixel temporal recognition, and compact edge-deployable networks, indicating that the central research question is no longer whether speckles contain usable information, but how that information should be encoded, sampled, and learned for each regime.