Image Processing And AI

I should be editing but I've been thinking about this for some time and wanted to see if I could peak someones interest with more skill than me. I'm exposed to Microsoft Azure/AI and Microsoft Flow a great deal day to day. Some of these tools would be very interesting for someone who is more a developer rather than a configuration wonk like I am and they are cheap (free for limited API hits) while MS is figuring it all out.

In short you can submit images to MS's Computer Vision AI and get it to suggest tags.

Try it here:
azure.microsoft.com/en-us/services/cogni...ces/computer-vision/
Maybe with these image URL's:
www.projectsuperwoman.com/stills/105/1.jpg
www.projectsuperwoman.com/stills/105/3.jpg

Why is this interesting? I would subscribe to a feed in a heartbeat that could provide the days images from instagram/deviantart/twitter/tumblr of women with an "S" or "WW" shield shirt/cosplay. I can follow the #supergirl tag - sure - but there is a lot of extra garbage in it. This tool could also perhaps be used in tagging our own galleries.

With the custom service you can "train" the tool to look for the "S" shield in theory then filter by gender adult content - you can even do celebrity facial recognition and filter in or out MB if you wanted.

Even if not today - coming to a service soon. Wow/Semi-Scary.

That's certainly an interesting idea they have going. Can definitely see the possibilities.

Will check it out Thanks Random

Peace.

/K

I think they and/or google also have a service that analyzes faces in pictures and tells you the sentiment(s) that it recognizes. Like "70% anger, 20% calmness, 10% love" or something like that. They have an example here: azure.microsoft.com/en-us/services/cognitive-services/emotion/

Incorporating this, you could probably tell apart "actual" supergirl content from "okay this is just some cute instagram girl wearing a tank top with an S on the chest".

Esteban wrote: "okay this is just some cute instagram girl wearing a tank top with an S on the chest".

Not that there is anything wrong with that. I like the idea of how to use emotion filtering - it also does "adult content" identification which means SFW vs NSFW content could be parsed.

One thing I've not been able to figure out if it can do is find if the image has been "seen" before to prevent duplicates from being put into a gallery. Sometimes the cute girl in a tank top gets re-shared more than a few times.

Also, good tip on Google's service - they seem like they are farther along on being able to tag content based on seeing something like an "S" crest. cloud.google.com/vision/

Okay, I guess we'll have to agree to disagree on wether or not cute Instagram girls should be included in a feed of that kind - so, this should be user customizable
I don't think those platforms work in a way that allows them to tell you something like "I've seen this one before". You send a picture and it tells you things about it - I think the duplicate detection has to be implemented by the consumer of the service. Also, where would the line have to be drawn? What about the girl in a sexy tank top and then the same picture, but with a caption or speech bubbles? You could store a hash of the binary file content of each image and use that for comparison, but you wouldn't detect to images as the same when one is a jpeg and the other is a png, or has been pulled off somewhere else via screenshot.

Gamification could be a different approach to this: Earn points and badges by flagging pictures as "I've seen this one before". User has to provide link to the original image in our feed for comparison. But maybe this would all be a great deal of over-engineering for a problem that doesn't annoy many people... at least to my knowledge