Challenge
Given 64 semi-finished Bayer color filter images, which have only Green cells filled, we should measure the dimension of the filled central cell. The problem has the following challenges:
- Highly noisy images
- Low texture/contrast
- High inter-class similarity
- Limited in quantity
Solution
Noise reduction technique was used to smooth the images. Unet deep learning then is applied to detect objects. Finally, draw and measure the bounding box of the detected objects.
- Unet works even in case the images have very bad quality
- Unet gives a high accuracy
- Unet requires fewer number of samples for training and testing data comparing with other models
Outcome
We were able to draw Larger and Smaller boundaries of the central cells at an accuracy that is equivalent to manual measurement done by experts.
