Fig. 2: Making uneven illumination EVEN: from the preliminary Evaluation of experimental artifacts to the automatic Enhancement of experimental images.

a Step 1: Definition of the evaluation criteria. We select quantitative metrics to detect vignetting in single fields of view and the mosaic effect in large images composed by multiple tiles: edge energy ratio \({E}_{{{\rm{edge}}}}\) and positive prominence \({P}_{+}\). The metrics are assessed manually on semi-synthetic images and on experimental measurements. b Step 2: Automatization and interpretation. The identification of uneven illumination is automatized by training a Linear Discriminant Analysis (LDA) model to differentiate images with and without uneven illumination. The quality metrics are used as features to train the model. The coefficients of the trained model can be analysed to investigate how the metrics are utilized to determine the presence of uneven illumination, while the decision score can be interpreted as a quality score to build a quality ranking of the input images. c Step 3: Evaluation and Enhancement of unseen images. Multimodal images are automatically optimized by exploiting the LDA prediction. The single channels (Ch. 1, 2, 3) of the raw input image are corrected independently with multiple correction methods, then they are provided as input to predict image rankings for the raw and corrected versions of each channel. The top image for each channel is selected automatically to generate an optimized multimodal image.