Learning illumination patterns improves single pixel classification accuracy
LIPS-PC: Learning Illumination Patterns for Single-Pixel Classification
Computer Vision and Pattern RecognitionMachine Learning
Summary
Single-pixel imaging captures objects using patterns of light and records just one number per pattern, making it slow and costly. This paper introduces a method where these lighting patterns are automatically learned to better identify object classes, rather than reconstruct the full image. The authors test their approach on handwritten digits, clothing images, and industrial X-ray weld images, showing it achieves better accuracy using fewer measurements compared to fixed patterns. This means fewer measurements are needed to classify objects, which is valuable especially when data is limited or costly to obtain.
What this means in practice
- •For industrial radiography teams: Improve defect detection accuracy in X-ray weld inspections by optimizing illumination patterns for efficient classification with fewer measurements.
- •For medical imaging engineers: Reduce dose and acquisition time in ionizing radiation imaging by learning measurement patterns targeted to classification tasks rather than full image reconstruction.
Authors
Tomer Porat, Talya Eden, Eliahu Cohen
Abstract
In single-pixel (ghost) imaging, known illumination patterns probe an object and a detector with no spatial resolution returns one number per pattern. Each measurement costs acquisition time and photons and, for ionizing probes, contributes to delivered dose. Motivated by automated radiographic defect screening, where a class label suffices, we introduce Learning Illumination Patterns for Single-Pixel Classification (LIPS-PC). The patterns are learned end-to-end while the measurement operator remains one fixed linear matrix, identical for every input. We parameterize it as a residual correction to a two-dimensional discrete cosine transform (2D-DCT), $D=D_{\mathrm{base}}+Δ$, and jointly train the correction and a compact classifier using cross-entropy, without a reconstruction objective. This yields a task-driven linear embedding through optimization. We evaluate LIPS-PC numerically on MNIST and Fashion-MNIST under a controlled protocol, then on RIAWELC, a real industrial X-ray weld-radiograph dataset. Across five budgets, $M\in{8,16,32,64,128}$, corresponding to sampling ratios from below 1% to 12.5%, the learned operator outperforms the best tested fixed transform by 9-15 percentage points at $M=8$ on all three datasets. Its eight-measurement accuracy requires roughly 16-32 measurements with the best fixed transform, indicating a $2\times$ to $4\times$ saving at comparable accuracy. The advantage generally shrinks as $M$ increases, with a small reversal on RIAWELC at higher budgets. These results show that learning what to measure is especially valuable when measurements are scarce.