Algorithm Results

Results

The results of each step of the image processing can be shown below.

Bilayer results

The microscope image of a bilayer flake, at 20x magnification.
The contours and bounding rectangles found by edge detection.
The bounding rectangles after size filtration.
The cropped rectangles for each flake.
Each flake after the bilayer mask.

The useful bilayer flake that we are looking for in this image is flake 3, which was correctly identified by the algorithm. Flake 6, however, is noise on the image that was flagged as a flake. We attribute this to the background color of the image, which is slightly corrupted due to light inconsistencies and hum bars, allowing this corner flake to fall into the bilayer mask. While not ideal, we determined that false positives are better than having missed detections and possibly missing useful flakes.

Other bilayer flakes that were tested were also correctly identified by the algorithm.

Monolayer Results

The microscope image of a monolayer flake, at 20x magnification.
The contours and bounding rectangles found by edge detection.
The bounding rectangles after size filtration.
The cropped rectangles for each flake.
Each flake after the monolayer mask.

The flake of interest here is flake 3. This flake was correctly identified as a monolayer flake by the algorithm. The gray dot, representing the center of mass of the white pixels, was used to determine what is versus is not a useful flake. Since the white pixels in flake 3 are centered around the center of mass, this was identified as a flake. Flakes 0, 1, 2 and 4, however, have white pixels that are on the exterior of the flake, which makes the center of mass on the black pixels. In this way, these flakes were not detected as flakes.

An issue that arose with monolayer flakes is the very slight contrast between the monolayer flake and the background. Given that the flake is only one atom thin, sometimes the flakes are hard to detect, even sometimes hard to see with the human eye. This issue was corrected with fine-tuning of the dilation during the edge finding step of the algorithm.

Few layer results

The microscope image of a fewlayer flake, at 20x magnification.
The contours and bounding rectangles found by edge detection.
The bounding rectangles after size filtration.
The cropped rectangles for each flake.
Each flake after the monolayer mask.

Flake number 2 is the few-layer flake of interest. This was correctly detected and identified by the few-layer mask, as shown by the white pixels. Since the few-layer flakes have the most contrast between them and the background, the results are generally the least noisy regarding the background.

Flake number 4, however, is not necessarily the few-layer flake of interest but was detected as a few-layer flake. This flake certainly falls within the hsv range for few-layer flakes, just by the naked eye, so this is a natural consequence of our algorithm.