A Variational Method for Dejittering Large Fluorescence Line Scanner Images

Abstract

We propose a variational method dedicated to jitter correction of large fluorescence scanner images. Our method consists in estimating a dense displacement field representing the jitter. An energy functional composed of a re-alignment criterion and a differential-based regularizer is especially designed. We adopt the half-quadratic splitting approach to decouple convex and non-convex terms in such a way that the original minimization problem can be solved by alternating two optimization steps -one using a proximal algorithm for the convex part and the other using an exhaustive search for the non-convex part. Experiment results on artificial and large real fluorescence images demonstrate that our method is not only capable to handle large displacements but is also efficient in terms of sub-pixel precision without modifying image intensities when comparing with image regularization-based approaches.

A Fast Dejittering Approach for Line Scanning Microscopy

“…Setting q = 1 for the DP algorithm, we used the two proposed approaches and the method from [10] to dejitter the simulated shifted images. The results can be compared qualitatively in Fig 2, and quantitatively in Tables 1 and 2 where we report the reconstruction SNR, the computing times, and the SSIM for completeness.…”

“…In this section, we first review the approach proposed in [10] and then introduce our contributions.…”

“…In [10], the authors propose to recover the displacement d by minimizing an energy inspired by the field of image registration. They suggest to solve:…”

“…We propose two different variational approaches to solve this problem. Both of them provide globally optimal solutions contrarily to [10]. They provide satisfactory results in short computing times, compatible with the huge dimensions of the InnoQuant system.…”

“…• The images can be huge, making it necessary to derive linear time algorithms. To the best of our knowledge, the only reference dealing with all the above specificities is a recent paper by Nguyen et al [10], which will be described later but suffers from significant limitations.…”