Optics-free imaging of complex, non-sparse and color QR-codes with deep neural networks

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional imaging systems rely on complex optical components for image formation.
• Quick Response (QR) codes are ubiquitous in modern applications, requiring reliable decoding.
• Developing compact and cost-effective imaging solutions is a persistent challenge.

Purpose of the Study:

• To demonstrate an optics-free imaging technique for QR codes.
• To utilize artificial neural networks (ANNs) for interpreting raw image sensor data.
• To assess the robustness of the optics-free approach under varying conditions.

Main Methods:

• Employing a bare image sensor without lenses or optics.
• Training artificial neural networks (ANNs) to reconstruct QR code images from raw sensor data.
• Experimentally evaluating system performance with variations in sensor-QR code gap (1mm, 5mm, 10mm) and alignment (translation, rotation).

Main Results:

• Successful optics-free imaging of both color and monochrome QR codes was achieved.
• The ANN demonstrated robustness against perturbations in gap distance and QR code alignment.
• Raw sensor data was effectively translated into human-interpretable visualizations.

Conclusions:

• Optics-free imaging using ANNs and bare image sensors is feasible for complex objects like QR codes.
• This approach offers a pathway towards simplified, non-anthropocentric camera designs.
• Potential applications include specialized, compact imaging systems where traditional optics are prohibitive.