*Example.*
So the final image size for the left screen half (in 1080p) in this
cartoonized-vectorized form is 40.1 KB. Using my new HUF1 (raw Huffman) format,
of course. :)
*Explanation.*
*THE** **GAZELLE** **DATA** **WRANGLER*
When an image is to be recognized, the "data wrangler" is Gazelle 22's first
line of attack. We start with the raw true-color (24 bit) RGB image.
"Data wrangling" typically includes the following operations on this image:
1. Blurring (optional)
2. Drastically reducing the number of colors (from 16 million to e.g. 64)
3. Finding all "regions"
[A "region" is a maximal connected set of pixels of uniform color.]
4. Vectorization (detecting straight lines)
--> Vectorized image can be saved too.
5. Compression (optional) - deleting some regions to save space
6. Image infilling (optional) - guessing pixels we dropped in step 4 or 5
[Because our current vectorization isn't perfect and can leave 1 or 2
pixel
wide gaps between regions.]
The list of pixels for each region is sent to the next recognition stage, for
example a cache of known region shapes.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T2be76bad62b78f24-Mba14b24c69de843636c1c9a7
Delivery options: https://agi.topicbox.com/groups/agi/subscription
