Imperfect 'E' in 'PERFECT'
Missing 'e' in 'Ingredients'
Invalid QR code
Interesting Units
We don't want any dark spots in your packages
'W' in flower

Flaw Detector

What does it do

It detects any flaws in the packages in packaging line and reports it. The flaw that it can identify are misplaced, missing or extra characters, any dark spots packages, mistakes in QR codes, etc.

How we built it

We followed following steps in our algorithm to acheive the results:

Aligned images with reference to a base image and then found the mean of large random subset of images(our case: 400 images)
We cropped out the black background out of image to reduce the computation & noise and used it to make a mask which is then further shrinked to remove noise due to missalignment of edges in images.
then, We calculate the weight matrix which assigns higher score to stable regions in the image and lower score to unstable regions(due to misalignment problems, Eg.the BOBST logo, edges of package)
Using the above set of precalculated mean, mask and weight matrix, we calculate the squares of the difference between the mean image and test image. This matrix essentially represents the difference(noise) in the image.
If the above calculated the noise is greater than the threshold, the image is classified as a image of flawed packaging.

Challenges we ran into

The noise added due to changing number at the button of the image (packaging), misaliged BOBST logo, etc was a causing a hurdle in our project
Solution: We overcome this issue by using the mask and weight matrix.

Accomplishments that we're proud of

Project: Our code successfully detects not just the misalgned, misplaced, missing characters but also defects like dark spots, torn out packaging, and improper color.
Multiple Implementaion Our team did pair programming and parallel programming to deliver successful implementaion in two languages (Matab & C++)
Team: Culturally diverse team with team members from 3 different countries and 3 different universities worked together to successfully complete the project