Computational Optics for Mobile Terminals in Mass Production.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning for Computer Vision

Background:

• Optical aberrations and manufacturing deviations degrade camera image quality, particularly in mass-produced mobile devices.
• Existing methods struggle to generalize correction techniques to stochastic manufacturing variations.
• Bridging optical design, machining, and post-processing is crucial for advanced imaging systems.

Purpose of the Study:

• To systematically model the relationship between lens system parameters and spatial frequency response (SFR).
• To develop an optimization framework for creating proxy cameras from machining samples.
• To propose a robust post-processing method for correcting optical degradation and manufacturing biases.

Main Methods:

• Constructed a perturbed lens system model linking deviated parameters to SFR.
• Developed an optimization framework to build proxy cameras using machining samples' SFRs.
• Proposed and implemented a dilated Omni-dimensional dynamic convolution (DOConv) for aberration correction.

Main Results:

• The optimization framework accurately constructs proxy cameras.
• The DOConv model effectively corrects optical aberrations and adapts to manufacturing deviations.
• Extensive experiments validated the method's performance on representative devices.

Conclusions:

• The proposed method successfully bridges optical design, machining, and post-processing.
• The DOConv approach enables robust computational photography, addressing real-world camera imperfections.
• This work advances the joint optimization of image signal reception and processing.