2.2.2. Determination of Particle Size Distribution by Microscopy-Assisted Digital Image Analysis

A total of 50 μL of distilled water was dropped onto the microscope slide, and 5 μL of the emulsion was spread over the slide with a syringe needle. It was carefully covered with a coverslip, and then immersion oil was dripped onto the coverslip. For imaging, 100×/1.25 oil, 160/0.17 objective lenses were used. Micrographs were obtained by a compound microscope (M83EZ, OMAX Microscopes, Kent, WA, USA) combined with a 5-megapixel CMOS camera (A3550U, OMAX Microscopes, Kent, WA, USA). The camera was located at the trinocular head, and micrographs were taken under transparent light conditions. At least 100 photographs were taken from 5 separate slides for each sample. The analysis was carried out immediately after the emulsion preparation and simultaneously

with the other analyses. The light intensity, field, and condenser diaphragm apertures of the microscope were adjusted carefully for reproducible results. No pre-correction was applied to the images. ImageJ/Fiji (ver.1.53c.) was used as the software, and Trainable Weka Segmentation (TWS) (ver.3.2.35) was used for the segmentation of the images.

For the training of TWS, 10 images with different droplet densities were selected. As these images were used in the training, they were not included in the droplet particle size analysis. Since the processing of individual images slows the training process, they were converted into a stack of a single image and then used in the training (Image > Stacks > Images to Stack). Then, 800 × 800 sized regions were duplicated from the center of these images (Figure 1). Afterwards, TWS was activated from the plugins (Plugins > Segmentation > Trainable Weka Segmentation). Droplets and background were drawn manually on each image and defined as different "Classes". After the first training, minor corrections were made to the images based on the responses obtained from TWS, and the training process was repeated until the results for the images were obtained from TWS (Figure 2). After training of TWS, segmentations of the micrographs of the samples were performed using the trained classifier on TWS. The images obtained as a result of TWS segmentation were turned into a single stack, and further operations were performed on this stack. Firstly, this stack was converted to 8 bits and then transformed into the binary format (Image > Type > 8-Bit, Process > Binary > Make Binary). After the empty droplets were filled with the "Fill holes" command, the "Open" command was applied to clear the pixels from the droplets (Process > Binary > Fill Holes-Open). Eventually, the particle sizes of these images were calculated (Analyze > Analyze particles).

**Figure 1.** The micrograms used in training of TWS.

**Figure 2.** The classified images after the training of TWS.

The calculation was applied to objects with a circularity value greater than 0.85 to exclude composite or half-droplet images, which are illustrated in the regions marked in red circles in Figure 3. A scale of 1 mm (100 × 10 μm) was used for the size calibration. The top of the scale was covered with a coverslip, and then immersion oil was dripped onto it. The number of pixels corresponding to the distance between two points with known distances was determined on the micrograph of the calibration slide. The measured distance was defined as global scale in ImageJ/Fiji (Analyze > Set Scale).

**Figure 3.** An example of droplet size analysis with classified images. While the particles circled in yellow are defined as droplets and used in the droplet size distribution calculations, those in the red circles, which are have a roundness value below 0.85, are not included in calculations.
