OCR used for vehicle plates, and guns

[fmnjr]

Just started hearing about this, maybe old news to others:

Google And Facebook Are Making a Photo Gun Registry Complete With Serial Numbers

Washington, DC — On Wednesday, The Firearm Blog posted a story that will likely blow most gun owners away. Earlier... View Article

www.secondamendmentdaily.com

 

[AvidIndoorsman]

 

[AvidIndoorsman]

Try this, take a screen shot of a gun on gun broker, and then use google search to do a reverse image search. It won't work. The algorithm requires it to be the exactly the same size, if you download the picture and then search it, you may get it to turn up.

Websites with complicated file structures blow up the process.

Think about this, law firms, O&G companies, medical companies, etc have hundreds of thousands of paper documents. These companies spend literally millions of dollars on OCR campaigns. Do you think they would do that if all they had to do was scan all their docs as .jpegs and then uploaded them into a web directory, and then could OCR search them via google?

I've managed OCR campaigns for 2 Oil and Gas companies, and actually did a podcast on the topic for the Oil and Gas Journal. Believe me if this was a real thing it would be incredible.

 

[MTGomer]

Try this, take a screen shot of a gun on gun broker, and then use google search to do a reverse image search. It won't work. The algorithm requires it to be the exactly the same size, if you download the picture and then search it, you may get it to turn up.

Websites with complicated file structures blow up the process.

Think about this, law firms, O&G companies, medical companies, etc have hundreds of thousands of paper documents. These companies spend literally millions of dollars on OCR campaigns. Do you think they would do that if all they had to do was scan all their docs as .jpegs and then uploaded them into a web directory, and then could OCR search them via google?

I've managed OCR campaigns for 2 Oil and Gas companies, and actually did a podcast on the topic for the Oil and Gas Journal. Believe me if this was a real thing it would be incredible.

This is WAY out of my area of expertise but we had a presentation a few months ago from an AI tech group that mostly caters to the oil and gas industry for cataloging leasing documents.
I wish I could remember their name.
They said we could upload all of our easement and lease agreement paperwork (100s of thousands of documents) and it was intuitive enough to catalog everything by section township and range as well by owner and is able to read and comprehend the language of the leases and compile The info, breaking down who we were leasing from, how much we owed them, when it was due and to set up reminders for us to make the payments, as an example of one of the things it could do.

 

[AvidIndoorsman]

This is WAY out of my area of expertise but we had a presentation a few months ago from an AI tech group that mostly caters to the oil and gas industry for cataloging leasing documents.
I wish I could remember their name.
They said we could upload all of our easement and lease agreement paperwork (100s of thousands of documents) and it was intuitive enough to catalog everything by section township and range as well by owner and is able to read and comprehend the language of the leases and compile The info, breaking down who we were leasing from, how much we owed them, when it was due and to set up reminders for us to make the payments, as an example of one of the things it could do.

Maybe Archeio --> sold to Quorum, who we use. Could also be Agile, now ThoughTrace, I've also worked with and/or had demos with Vanguard, efileCabinent, DRS, docuscan, Land Information Services... and then a couple of inhouse products built by various land service companies. The industry is contracting right now, efficiency is a giant buzz word and my junk mail is stuffed with emails from these companies claiming to have solved all our problems.

When push comes to shove these companies, can basically do what you described, but not without a lot of man power, especially if you data is dirty. If you have everything digital and already indexed by STR, Owner, Unit, Prospect, etc it's pretty quick to get them input. Provisioning is way more complicated, programs like Thoughttrace require an analyst to go through every lease and attribute all the provisions. If you have standard lease forms, then the machine learning can kick in and use a couple of different forms as templates and do it for you, if like any company operating in the US today, you have a mix of legacy agreements, assignments, no standard or perhaps even hand written in language the program blows up. Also if anyone tries to sell you on some automated GIS system it's all BS you can kinda make it work with huge tracts in a Jeffersonian state, but if you get near any developed area/ east texas/ or anywhere back east the systems are useless. Your best bet is a GIS tech who is good with COGO tools or getting something out of the box from a company like Whitestar.

I think if you are a wind or solar company starting out with clean data, and using standardized forms you can build a pretty efficient system with digital documentation. If you are an established company in a sector with legacy data of any kinda, there is no easy button you can click. You can arrive at the end point I state above "googling your docs" through a software like Dynamic Docs (Archeio) but lots of work is required.

Cause you know... these guys were super worried about database optimization.


Lol so my apologies to the OP, kinda an irritating topic for me right now not trying to poop in your Easter basket.

 

[Irrelevant]

This is WAY out of my area of expertise but we had a presentation a few months ago from an AI tech group that mostly caters to the oil and gas industry for cataloging leasing documents.
I wish I could remember their name.
They said we could upload all of our easement and lease agreement paperwork (100s of thousands of documents) and it was intuitive enough to catalog everything by section township and range as well by owner and is able to read and comprehend the language of the leases and compile The info, breaking down who we were leasing from, how much we owed them, when it was due and to set up reminders for us to make the payments, as an example of one of the things it could do.

Sounds like they were selling something... never trust a salesman.

 

[fmnjr]

Maybe Archeio --> sold to Quorum, who we use. Could also be Agile, now ThoughTrace, I've also worked with and/or had demos with Vanguard, efileCabinent, DRS, docuscan, Land Information Services... and then a couple of inhouse products built by various land service companies. The industry is contracting right now, efficiency is a giant buzz word and my junk mail is stuffed with emails from these companies claiming to have solved all our problems.

When push comes to shove these companies, can basically do what you described, but not without a lot of man power, especially if you data is dirty. If you have everything digital and already indexed by STR, Owner, Unit, Prospect, etc it's pretty quick to get them input. Provisioning is way more complicated, programs like Thoughttrace require an analyst to go through every lease and attribute all the provisions. If you have standard lease forms, then the machine learning can kick in and use a couple of different forms as templates and do it for you, if like any company operating in the US today, you have a mix of legacy agreements, assignments, no standard or perhaps even hand written in language the program blows up. Also if anyone tries to sell you on some automated GIS system it's all BS you can kinda make it work with huge tracts in a Jeffersonian state, but if you get near any developed area/ east texas/ or anywhere back east the systems are useless. Your best bet is a GIS tech who is good with COGO tools or getting something out of the box from a company like Whitestar.

I think if you are a wind or solar company starting out with clean data, and using standardized forms you can build a pretty efficient system with digital documentation. If you are an established company in a sector with legacy data of any kinda, there is no easy button you can click. You can arrive at the end point I state above "googling your docs" through a software like Dynamic Docs (Archeio) but lots of work is required.

Cause you know... these guys were super worried about database optimization.
View attachment 117940

Lol so my apologies to the OP, kinda an irritating topic for me right now not trying to poop in your Easter basket.

I like humor, even at my own expense. Not to go off on a tangent, but getting butt-hurt about people poking fun, or being sarcastic, well, that just proves one is a pansy. Which we have way too much of nowadays.

Back to the topic of OCR, obviously I'm not claiming to be informed about it but with all the forms of photo recognition out there (and the furious pace at which it is developing) I don't think it's too far-fetched of a concept to grab info off photos on the net. As you so aptly point out, others are very informed about such things. I'm very aware there are some really smart people working on data mining in general and, if I've thought of it, they surely have and will somehow figure it out.

 

[AvidIndoorsman]

I don't think it's too far-fetched of a concept to grab info off photos on the net.

I agree with this... I think in practice your worst case scenario would be someone poaching, selling their weapon online, and the DW finding your listing and using it to prove you owned the gun in question that was used to poach.

Aspects of the article are true, uploading metadata exists on some platforms... I'm sure HT stores your IP with photo uploads. You can OCR from .jpegs and could pull a serial number off a high res photo. But circumventing the government to create a index of gun owners via uploaded photos? That would be a massive campaign with a huge capital spend, would require tech that is not currently available and wouldn't be particularly productive as most of those gun photos are from retailers not individuals, and even then it's only a tiny tiny fraction of all guns in the US... so it wouldn't be very helpful even if google wanted to big brother you.

 

[Ben Lamb]

40 years ago, we were never going to be able to watch movies and talk to people on our phones.

20 years ago we didn't think the computers would be able to tell us what we wanted to buy. Now, they know before we do.

10 years ago, we didn't think that vehicles would ever be able to drive themselves.

I'm keeping my tin foil hat on boys. I'm just ahead of the curve.

Molon Labe.

 

[fmnjr]

 

[AvidIndoorsman]

40 years ago, we were never going to be able to watch movies and talk to people on our phones.

20 years ago we didn't think the computers would be able to tell us what we wanted to buy. Now, they know before we do.

10 years ago, we didn't think that vehicles would ever be able to drive themselves.

I'm keeping my tin foil hat on boys. I'm just ahead of the curve.

Molon Labe.

I'm more putting this one in category of personal heads up displays, personal jet packs, real hover boards, flying cars, meals in a pill, and reliable public transportation in the United States... sure all of these are plausible but none of them are going to be a thing.

 

[MattK]

I use an OCR program to take financial statements (pieces of paper) and create spreadsheets out of them. If I have a good copy of a statement (clean), use 300 plus dpi to scan it, I get over 95% accuracy. If the scan is tilted, blurry, or I drop the scan below 300 dpi, I find about 60% accuracy which is unusable. Facebook is continuously working on algorithms and OCR programs to make sure "bad stuff" doesn't hit their pages but they still need people to verify the "bad stuff" because there are too many mistakes and too much slips through their processes. They won't give their accuracy rates but it's bad enough there are congressional hearings about it. As wllm1313 has pointed out, there is a long way for these programs to go and it would take a ton of capital that most companies aren't willing to expend for the results they will get (and the money they will save).

Another interesting angle to this is the expense reports for businesses. We are just starting to use an OCR program from a third party vendor that will allow a worker to take a picture of a receipt. The program then takes all the important information and puts it into a specific layout, sends it to the correct person to "ok" as an expense and file it into the correct accounts. It's also able to tell the person right away if it is a viable expense...weeds out any expense that is specifically not allowed so it doesn't waste manager's time denying it or making an incorrect assessment of the expense.

OCR has made some monster steps, just not the steps that some people believe it has.

 

[Ben Lamb]

I'm more putting this one in category of personal heads up displays, personal jet packs, real hover boards, flying cars, meals in a pill, and reliable public transportation in the United States... sure all of these are plausible but none of them are going to be a thing.

Said the Horse salesman to the horseless carriage.

 

[AvidIndoorsman]

Said the Horse salesman to the horseless carriage.

What's your best elevator speech about why I should invest in your "interweb gun recognition" idea, how am I making my return.

 

[LuketheDog]

"Every gun picture uploaded to Google, Facebook, Instagram, Youtube and Twitter is being automatically scanned using OCR and its serial number is now stored."

Yeah, unless they're sending little green men out to take a specific clear pic of the firearm's little, tiny serial number, I'm not buying what he's selling. Also, Jalopnik is not "a website for selling cars and car parts"... -2 for the author's credibility.

 

[Gr8bawana]

"Every gun picture uploaded to Google, Facebook, Instagram, Youtube and Twitter is being automatically scanned using OCR and its serial number is now stored."

Yeah, unless they're sending little green men out to take a specific clear pic of the firearm's little, tiny serial number, I'm not buying what he's selling. Also, Jalopnik is not "a website for selling cars and car parts"... -2 for the author's credibility.

 

[MTGomer]

Like most things, I suspect the truth is found between
“the government is compiling a gun registry based on serial numbers from Facebook photos”
and
“e-bike technology will never evolve beyond where it is today”


Aka: in the middle, where logic resides.