123
-=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- (c) WidthPadding Industries 1987 0|380|0 -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=- -=+=-
Socoder -> C/C++/C#/Other -> python graphics

Posted : Monday, 11 October 2010, 01:35
kenkit
"Thanks i'll try CSI but I run onto soar already the only problem is that it's in JAVA

Posted : Monday, 11 October 2010, 01:41
Afr0
You mean like in CSI?
I've been wondering about that myself.
I think it's terribly difficult, and I can tell you Python isn't the language to do it in.

-=-=-
Afr0 Games

Project Dollhouse on Github - Please fork!
Posted : Monday, 11 October 2010, 06:07
JL235
Afr0 and I can tell you Python isn't the language to do it in

Why?

Kenkit anybody who knows how to create image detection software something like compare ?

I can't tell you how to do this, but I can point in the right direction. Many implementations use an Artificial Neural Network where you have several layers of simple 'neurons'. The first layer takes inputs from you (the photo) and gives out inputs to the next. This continues for each layer until you reach a layer of outputs where you get your results. The neurons are connected by weighted connections that alter the input as it's passed from one neuron to the next. The idea is that by altering these weights you can alter how they will interpret a given input.

You get them working using test data. Depending on if the output is correct or incorrect you alter the connections internally (if the output is correct they are strengthened, if incorrect then they are weakened). The theory is that over time the neural network will 'learn' how to search for the data your inputting.

However they can learn the wrong things. There was a neural network built to spot if a tank was an American tank or a Soviet tank. The Soviet pictures were of much lower quality then the US ones so it just learned to spot how grainy the picture was, and gave an output based on that. This is one of _many_ stories I've heard of these messing up.

If you Google around you can find lots of sites with peoples dissitations and other academic papers. I'd search through them for topics like 'neural networks for image detection' or similar, and you'll probably find several techniques for building this.

I hope that helps!

-=-=-
PlayMyCode.com - build and play in your browser, Blog, Twitter.
Posted : Monday, 11 October 2010, 08:52
Afr0
Why?


Because it's a scripting language, first and foremost. I'm pretty sure it wasn't designed to handle use-cases such as this. Frankly I'd go with C or C++ for this. Or maybe even Assembly, depending on the wanted speed.

-=-=-
Afr0 Games

Project Dollhouse on Github - Please fork!
Posted : Monday, 11 October 2010, 09:09
JL235
If I write a really slow algorithm, but do it in C, it's gonna be dead slow. If I write the same in Python, but use a really fast algorithm, it'll be fast.

If Pythons overhead does become a problem then he could move sections of it out into C code then. There are several Python libraries available for helping users to mix C and Python code.

-=-=-
PlayMyCode.com - build and play in your browser, Blog, Twitter.