HugoRune's profile - activity

edit after 5 years:

i marked Michael Burdinov's answer as the best answer. Because i have not got any answer better than his answer yet.feel free to give a better answer if you have.

Given an image containing a mirror symmetrical object, such as a car or a butterfly, I want to find the symmetry axis:

Due to perspective distortions the two halves of a symmetrical object will never match exactly. The background and noise further complicate things.

But given the assumption that the image of the object is mostly mirror symmetric and takes up a large portion of the picture, it should be possible to find a best fit for the axis of symmetry.

How can I find this axis?

I tried flipping the input image, then executing SURF and FindHomography on the original and the flipped image, but the results are very poor, and not robust at all.

Is there a better way?

Is there a way to apply a blur or median smoothing filter to an image, while supplying a mask of pixels that should be ignored?

I have a height map from a laser-scanner which I want to smooth. The map is not continuous; wherever the laser was not reflected, the map simply contains no height data.

If I arbitrarily set the height for missing values to zero (or any other value) and then blur the image, this will introduce a lot of error around these missing values, and around all edges and holes of objects.

So I need to supply a binary mask to the blur. Masked pixels should be ignored when calculating the new blur or median value of neighbor pixels.

How can I accomplish this?

I want to detect edges in a range image. The image contains distance values from 0 meters to 2,5 meters as 32 floats.

Canny edge detection works very well for my images. However the cvCanny method only accepts single-channel 8bit input images.

Converting the distances to 8bit results in a resolution of about 1cm, which is far too low. The original images have a resolution of below 1mm

I can generate different 8bit images that only cover a small part of the full range each, and run canny on all of them, but that is slow and cumbersome, and introduces artifact edges.

Is there a way to run canny edge detection on the full range of the input data?

The speed is not an issue compared to all the other stuff that needs to run, and this method happens to match well with the form of noise I am dealing with: There are a couple highly localized noise peaks, one or two pixels wide, that disturb normalization; and those are removed by erosion

unxnut's suggestions is probably the right way to go at this, but I found a quick hack that gets me mostly what I want:

To stretch an image so the topmost few pixels are clipped:

• cvErode srcImage n times --> erodedImg // this will remove all the highest localized peaks
• cvMinMaxLoc erodedImg // this will get the maximum value
• Scale srcImg * (255.0 / maxValue) // this will set the maximum value of the eroded image to 255 and clip all higher values

removing low peaks works the same, but was not needed in my case

To contrast-stretch an image I use

CvNormalize(img,img,0,255, NormType.MinMax);

This will set the minimum to 0 and the maximum to 255, stretching all values in between.

But is there a way to specify a percentage of values that should be cut off? If I want to stretch the image so that at least 5% of the resulting array are zeroes and 5% of the resulting array are 255,
(Or alternatively, stretch the image so that 5% of the lowest values and 5% of the highest values are clipped),
what combination of opencv commands could I use to accomplish that?

Yes, this helps, thanks. My own searches all came up empty because I was searching for an implementation using openCV, but I guess converting these ones to use the openCV datatypes and functions will be a lot easier than converting the native c canny code.

Instructions on how to download and install the prebuilt windows dlls can be found at
http://docs.opencv.org/doc/tutorials/introduction/windows_install/windows_install.html#windows-install-prebuild

It is basically just downloading and executing the first link from the download page

I want to try a few things with a modified canny algorithm, in order to adapt it to detect centerlines instead of edges. In particular i want to switch the standard sobel filter to detect edge direction.

To this end I am looking for an implementation of canny that I can modify.

However the source for the OpenCV Canny implementation is very hard to translate into a high-level language. I am currently using a C# wrapper for openCV, and cannot use pointer arithmetic, memset etc. If possible I would like to keep all my code in one language.

Is there an implementation of canny using the basic openCV functions and classes that would be easier to translate? Something in Python would probably work just fine.

the canny algorithm uses the sobel operator to calculate the gradient direction atan2(sobelY,sobelX). I am not sure what you are asking, could you clarify your question?

I added some ideas to my answer