Anonymous user
Talk:Hough transform: Difference between revisions
→Can I get some pseudo-code...and possibly an amen
Line 45:
::::: You have my sympathy, but I question the extent to which things must be identical. The point is that the generated output is not strictly an image, but rather a graphical plot of intensity over Hough Space. This means that you can scale either axis or rotate the angular axis and still get something that is “the same”. Add in the point that different languages have different natural interpretations of pixels (How many color channels? Over what range?) and you really end up with something other than what you might wish. However, I ''specifically'' want to permit multiple methods of implementing the transform; if someone's got some clever technique or useful library, let them use it! This is not a task that calls for slavish copying. As long as the mathematical transform is implemented and it is capable of processing images, I would count it as a solution (and I'm deliberately vague about whether it should be a color or B+W image).
::::: For the record, it's not very difficult to implement from scratch IMO once you get the idea of what the transform is doing. –[[User:Dkf|Donal Fellows]] 22:55, 9 August 2010 (UTC)
:::::: Yeah, it's not difficult to implement from scratch... the problem is there are many algorithms that perform this transform...just like almost every mathematical transform. And, in cases like the DFTF, there is an established algorithm that is generally used because it is the fastest, or most precise, etc whereas the others that perform the same task are disused. This is the problem I had. It was no problem for me to design the algorithm...but I ended up doing it 4 different ways in an attempt to figure out the best way. This is why I think we should define a specific algorithm that performs this transform. Leaving the option for varied inputs and outputs is cool, but if we want to be able to compare implementations something in this task should be specified. Especially, the inputs and the outputs. I can't even tell if the TCL output is wrong. (though I have the feeling it is, by definition of the hough transform you have all the information within one period of cos and sin by plotting two periods you are simply replotting information you already know from the first period. But that is for another thread.)
:::::: Anywho if I'm the minority vote here, I am not going to go cavalier and modify the task spec. But, I think the task designer should ''seriously'' consider making the task spec more specific. If not for the sake of clarity, then for the sake of the sanity of non-guru programmers and people not willing to do serious research on the topic. --[[User:Cferri|Chris Ferri]]
==PNG Image==
| |||