image processing of food labels
Post on 12-Feb-2017
201 Views
Preview:
TRANSCRIPT
{saturated fat: 6.7g,added sugar: 8.1g,
fibre: 8.5g,proteins: 9.9g,
salt: 1.4g}
1estambolieva@gmail.com
Practical Image Processing
Social Entrepreneur
2estambolieva@gmail.com
Huge Machine LearningEnthusiast
ME
3estambolieva@gmail.com
SUGARWISE
EXCEEDS
your daily allowance
of added sugar
try instead
4estambolieva@gmail.com
THE PROBLEM
5estambolieva@gmail.com
TESSERACT
https://github.com/tesseract-ocr
6estambolieva@gmail.com
CONVERT TO GRAYSCALE
python: PIL libravy
7estambolieva@gmail.com
PROBLEM: BINARIZATION
python: PIL library does not work here
8estambolieva@gmail.com
colour histogram
~ 500 000 pixel values, each pixel value is a different colour
9estambolieva@gmail.com
k-means clustering
image source: Wikipedia
http://stackoverflow.com/questions/3241929/python-find-dominant-most-common-color-in-an-image
10estambolieva@gmail.com
kmeans: let's try it out
40x40 pixel window
"first window"
11estambolieva@gmail.com
kmeans: let's try it out
40x40 pixel window
"first window"
154
12estambolieva@gmail.com
kmeans: let's try it out
40x40 pixel window
"first window"
13estambolieva@gmail.com
JPG v.s. PNG
40x40 pixel window
14estambolieva@gmail.com
kmeans: results on full image
15estambolieva@gmail.com
SOLUTION: sloppy way
colour (pixel values) histogram
find middle pixel valueeverything below it goes blackeverything above it goes white
16estambolieva@gmail.com
SOLUTION: sloppy way
17estambolieva@gmail.com
side to side comparison
18estambolieva@gmail.com
LINE DELETION
work with windows or full image?
remove black regions > 400 pixels?
remove uninterrupted blackregions?
Are we missing something?
19estambolieva@gmail.com
black regions > 400 pixels
window-wise
20estambolieva@gmail.com
uninterrupted black regions
window-wise
21estambolieva@gmail.com
uninterrupted black regions
window-wise
22estambolieva@gmail.com
SOLUTION: line detection
image-wise
23estambolieva@gmail.com
SOLUTION: line detection
window-wise
24estambolieva@gmail.com
line detection: untested
image-wise
fourier analysis
25estambolieva@gmail.com
bonus: MIN FILTER
min filter PIL default: 3 pixels
26estambolieva@gmail.com
bonus: SKELETON
python: openCV
27estambolieva@gmail.com
CHALLENGE: WHITE PIXELS INLETTERS
28estambolieva@gmail.com
THANKS!
top related