Lensless inference camera: incoherent object recognition through a thin mask with LBP map generation.

Area of Science:

• Computer Vision
• Optical Engineering
• Machine Learning

Background:

• Traditional object recognition systems often rely on computationally intensive image reconstruction.
• Lensless imaging offers potential for reduced hardware complexity and cost.
• Real-time processing remains a challenge for many optical inference systems.

Purpose of the Study:

• To propose and evaluate a lensless inference camera (LLI camera) for efficient object recognition.
• To introduce a novel preprocessing technique for optically encoded patterns.
• To demonstrate the practical applicability of the LLI camera under varying environmental conditions.

Main Methods:

• Development of a lensless inference camera utilizing a mask for optical encoding.
• Implementation of a new preprocessing approach: local binary patterns map generation.
• Optical experiments for handwritten digit recognition and gender estimation.

Main Results:

• The LLI camera achieves real-time inference by performing efficient preprocessing instead of computationally expensive reconstruction.
• Local binary patterns map generation enhances the robustness of encoded patterns to disturbances.
• Successful demonstration of handwritten digit recognition and gender estimation under challenging conditions.

Conclusions:

• The proposed LLI camera offers a computationally efficient solution for real-time object recognition.
• The novel preprocessing method significantly improves system reliability in practical scenarios.
• Lensless imaging combined with efficient preprocessing is a viable approach for advanced machine vision applications.