Learning to Estimate Kernel Scale and Orientation of Defocus Blur with Asymmetric Coded Aperture

TL;DR

This paper introduces a deep learning framework that estimates the kernel scale and orientation of defocus blur using synthetic data and 3D ConvNets to enable rapid lens focus adjustment in machine vision systems.

Contribution

It presents a novel deep learning approach with synthetic data generation for estimating defocus kernel parameters, improving focus adjustment accuracy.

Findings

Effective estimation of defocus kernel parameters demonstrated

Synthetic asymmetric coded aperture images facilitate training

Framework improves focus adjustment speed and accuracy

Abstract

Consistent in-focus input imagery is an essential precondition for machine vision systems to perceive the dynamic environment. A defocus blur severely degrades the performance of vision systems. To tackle this problem, we propose a deep-learning-based framework estimating the kernel scale and orientation of the defocus blur to adjust lens focus rapidly. Our pipeline utilizes 3D ConvNet for a variable number of input hypotheses to select the optimal slice from the input stack. We use random shuffle and Gumbel-softmax to improve network performance. We also propose to generate synthetic defocused images with various asymmetric coded apertures to facilitate training. Experiments are conducted to demonstrate the effectiveness of our framework.