ADMIRE: a locally adaptive single-image, non-uniformity correction and denoising algorithm: application to uncooled IR camera

TL;DR

The proposed method uses an hybrid scheme including an automatic locally-adaptive contrast adjustment and a state-of-the-art image denoising method that permits to correct for a fully non-linear NU and the noise efficiently using only one image.

Abstract

We propose a new way to correct for the non-uniformity (NU) and the noise in uncooled infrared-type images. This method works on static images, needs no registration, no camera motion and no model for the non uniformity. The proposed method uses an hybrid scheme including an automatic locally-adaptive contrast adjustment and a state-of-the-art image denoising method. It permits to correct for a fully non-linear NU and the noise efficiently using only one image. We compared it with total variation on real raw and simulated NU infrared images. The strength of this approach lies in its simplicity, low computational cost. It needs no test-pattern or calibration and produces no "ghost-artefact".

Figures (15)

• Figure 1: On the left : an image (RAW) taken with an infrared camera. The non uniformity is so strong that it is hard to distinguish between the noise and the underlying landscape. On such an image performing an identification, matching pattern, etc. is almost hopeless. On the right the same image corrected with the proposed method.
• Figure 2: Two histograms $h_1$, $h_2$ (left side) and the corresponding midway histogram $h\_mid$ (on the right), compared to the direct histogram average, which would create two modes (centered at $n1$ and $n2$) and is therefore wrong. The uniform equalization would destroy the grey level dynamic and create artefacts. It is not a good candidate to get good quality images.
• Figure 3: On the left : an uncorrupted test image (boat). On the middle : the result of the MIRE algorithm ($s^*=0$); the produced image is the same. This experiment was done using ipol_mire. On the right : the result of the locally adaptive variant of MIRE described in section \ref{['sec:adaptive']} ($s^*=0$ everywhere in the image). As predicted the algorithms does not make the image worse or create artefacts (safety check). Results on real raw images corrupted with non uniformity are detailed in section \ref{['sec:experiments']}.
• Figure 4: On the left : a real raw image produced by an infrared camera. On the right : the result of the locally adaptive MIRE algorithm (see section \ref{['sec:adaptive']}). It is strongly corrupted by noise.
• Figure 5: On the left : a real raw image produced by an infrared camera. Middle : the result of the algorithm with the parameter $s=1$. On the right with $s=7.5$. For example, focusing on a zone in the middle of the image the image with $s=7.5$ is nicer but, the zone below poles is bad. On the other hand in the image processed with $s=1$ the zone in the middle is still corrupted by the non uniformity. Thus, a fixed $s$ parameter for the whole image will not lead to the best quality possible everywhere.