Color lens-free imaging using multi-wavelength illumination based phase retrieval.

Jasper Mariën,Chris Van Hoof,Richard Stahl,Abdulkadir Yurt,Andy Lambrechts

doi:10.1364/oe.402293

Abstract

Accurate image reconstruction in color lens-free imaging has proven challenging. The color image reconstruction of a sample is impacted not only by how strongly the illumination intensity is absorbed at a given spectral range, but also by the lack of phase information recorded on the image sensor. We present a compact and cost-effective approach of addressing the need for phase retrieval to enable robust color image reconstruction in lens-free imaging. The amplitude images obtained at transparent wavelength bands are used to estimate the phase in highly absorbed wavelength bands. The accurate phase information, obtained through our iterative algorithm, removes the color artefacts due to twin-image noise in the reconstructed image and improves image reconstruction quality to allow accurate color reconstruction. This could enable the technique to be applied for imaging of stained pathology slides, an important tool in medical diagnostics.

Highlights

In recent years, lens-free imaging (LFI) has evolved into a widely applied imaging technique, based on the principles of in-line holography [1,2]
CMOS image sensors are only sensitive to intensity of the incoming light and information on the phase delay is lost at the image sensor during hologram acquisition
We demonstrate a technique for robust image reconstruction in color lens free imaging

Summary

Introduction

Lens-free imaging (LFI) has evolved into a widely applied imaging technique, based on the principles of in-line holography [1,2]. CMOS image sensors are only sensitive to intensity of the incoming light and information on the phase delay is lost at the image sensor during hologram acquisition This results in the appearance of the object’s twin image after image reconstruction [14,15,16]. Important improvements in lens-free image reconstruction were obtained through the introduction of several deep-learning based techniques to perform phase retrieval, improve image resolution or use a deep-learning network to match lens-free image reconstructions with brightfield microscope images [25,26,27,28] Such approach requires careful training of the network but once trained, a deep-learning network performs its task accurately, and efficiently

Results

Discussion

Conclusion