Help removing/reducing noise in digitized data

Guillermo

2011 9 20

0 个回答

1 次查看（30 天）

0 个投票

Hello community,

I have question regarding a project I'm working on. I have recently digitized a series of pictures of an insect walking (digitized the ends of each leg), however ,due to being unable to click on the exact same spot every time (while using ginput) there is some noise in the data. I'm trying to create a routine that lets me know when each leg is moving and when it's not moving, however, this noise makes it difficult to do.

Is there a way of removing this noise, or a way of predicting where the leg should be next so that the noise doesn't have an effect?

9 个评论
显示 7更早的评论隐藏 7更早的评论

Guillermo 2011-9-21

Well, what i want to do is quantify how long the "not moving" (stance) phase of each leg (tripod corner in this example) lasts, and how long the "moving" (swing) phase of each leg is. Due to the fact that while manually digitizing the data,m y colleagues and i never clicked on the exact same point every time, i can't say if the movement = 0, then it's in stance, as there is movement due to the noise. Even if i average a set of frames, there will still be movement. So i guess what I'm asking for is how can i set the parameters so that it could discriminate between the slow moving swing phase and the noise of the stance phase? I have attempted setting manual parameters (e.g., if there is between 0-3 pixels of movement it's noise, if its more than 3 its an actual swing phase) however, at the beginning and end of each swing phase the movement is slow enough to fall within the stance phase parameter, and if i lower the parameters, the noise will be high enough to be considered a swing stance, giving me both, false positives and false negatives which, when looking at long bout lengths would inflate the stance phase, and make it seem longer than what it is.

Jan 2011-9-22

@Guillermo: The question is not precise as you can see by the best possible answer: "yes".

Creating a meaningful answer would be easier, if you concentrate on a specific problem. If you want to apply smoothing, that all details about the insect are confusing: You want to filter a matrix of X- and Y-values. If you want to do the marking or the foot automatically, then it would be helpful if you explain the available input, e.g. frequency, resolution and color depth of the images. And example image might be helpful also.

Guillermo 2011-9-22

@Jan Simon, I was originally interested in the first option, about smoothing, however, if it were possible to mark the foot automatically, that would make my life easier in the long run. The images we are currently analyzing are from a black and white video taken at a frequency of 10 frames/second, with a resolution of 1250 x 980. here's a zip file with some example images:

http://www.sendspace.com/file/92avwu

请先登录，再进行评论。

请先登录，再回答此问题。

Follow Question