Note that there are some explanatory texts on larger screens.

plurals
  1. PO
    text
    copied!<p>Here's some general high-level advice for how to get started.</p> <p>Basically, what you are doing is an optimization problem. These algorithms are used for a lot of problems, and there are several well-known ways to do this. They boil down to this</p> <ol> <li>Create a scoring function that can tell you a single number of how good a result you have. The bigger the number the better.</li> <li>Create a function that takes the input and some parameters and produces an output that can be scored</li> <li>This is important: The scoring function should be somewhat continuous based on the parameters to #2. If you had two parameters, and plotted it in 3D (param1, param2, score), it would look like a bumpy surface with big hills.</li> <li>Your job now is to find the maximum in the surface. You may have more than two parameters -- in that case, you have an N-D surface -- but the idea is the same</li> </ol> <p>Look up Hill-climbing, genetic algorithms, or optimization problems. A good python book with code is "Programming Collective Intelligence" by Toby Segaran.</p> <p>Generally hill-climbing is something like:</p> <ol> <li>Make a good guess of the parameters</li> <li>Create the output and score</li> <li>Change one parameter slightly</li> <li>Score the output</li> <li>If it's better keep going in this direction, if it's worse, change direction.</li> <li>If you are stuck -- go somewhere else in the surface and try there.</li> <li>If you find a local maximum, but it's not good enough -- go somewhere else and try there</li> </ol> <p>Look up the actual algorithms though, they are somewhat more complex than this.</p> <p>A lot of the research boils down to coming up with a good scoring function and a good way to know what parameters will work and how to use them.</p> <p>Using this general outline -- just try brightness/contrast as your output generating function (brightness and contrast are inputs). For scoring, you will need a way of comparing two photos for a match -- to start, pick something simple (perhaps hard-code an area to check). </p> <p>Once you get it going, you will have more insights into how to do this, and can go back to the papers for ideas.</p>
 

Querying!

 
Guidance

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload