I want to copy the part of an image which is descibed by a SURF descriptor. I know that the 9x9 filter in surf has a scale of 1.2. So if I have for example a descriptor with a scale of 1.2 is the part I can copy the 9x9 pixels around the descriptor point? And if I have a scale of 1.6 I would calculate the part with "9 / 1.2 * 1.6 = 12"? The subimage would be 12x12 pixels than?
On the other hand I read this in the OpenSURF documentation:
"The first step in extracting the SURF descriptor is to construct a square window around the interest point. This window contains the pixels which will form entries in the descriptor vector and is of size 20s, again where s refers to the detected scale."
So maybe I get the subimage if I take the 20*scale pixels around the descriptor point?
I don't know why it should be 20!? The first solution is more clear I guess...
Thanks!
According to SURF original paper, I can answer you with two things:
The first solution you mention is the way to generate the filter size for an octave in SURF. However we don't have filter size 12x12. We just have {9x9,15x15,21x21,27x27},{15x15,27x27,..,..},{27x27,..},... It is not the region around the SURF key point for extracting the descriptor.
The sub image that covers a descriptor region is rotated window (based on dominant orientation) with the size 20*scale around the descriptor point. In my opinion, practical experiments allow the authors to pick that number (20). Similar to minimum Hessian threshold for eliminating non-SURF key points, they are heuristic method.
If the implementation is exactly the one described in the original SURF paper, you have to use 20s. 9x9 is the starting filter size to detect interest points, but 20s is the window dimension to describe the area around an interest point.
Related
I am trying to erode objects in a binary image such that they do not become smaller than some fixed size. Consider, for instance, a binary map composed of connected components (blobs), wherein one defines blob size by either the minimal or maximal antipolar (anti-perimetric) distance (i.e., the distance between two points that are as far from one another as they can be on the perimeter or contour of the blob; if the contour consists of N consecutively numbered points, then the distances evaluated would be those between points 1 and N/2+1, points 2 and N/2+2, etc.). Given such an arrangement, I seek to erode these blobs until the distance metric reaches a specified limit. If the blobs were simple circles, then the effect could be realized by ultimate erosion followed by dilation to a fixed size; however, the contour of an irregular object would be lost by such a procedure. Is there a way to achieve such an effect for connected, irregular components using built-in functions in MATLAB?
With no image and already tried code presented I can understand you wrong, but may be iterative using bwmorph with 'thin','skel' or 'shrink' will help you.
while(cond < cond_threshold)
bw=bwmorph(bw,...,1); %one of the options above
cond = calc_cond(bw);
end
need some help here on image processing. I'm using Matlab and try to segment the following figure based on the two major peaks (in yellow). The color yellow means higher value and blue means low value (on z-axis, or image color from 0 to 1 for your convenience). The ideal cut is roughly the line from point (1,75) to (120,105). But I want a systematic way to derive this rather than by observation.
My intuition was to first identify the two peaks (based on this), and then classify each point/pixel on this figure to the two peaks (the metric here is to compute the shortest Euclidean distance to the edge of the two peaks).
And I end up with the following fig.
As you can see, the cut is pretty much a straight line, which I'm not quite satisfied. Maybe I can use the orientation of the peak circle and somehow tilt the line.. but I'm not sure how to do so? Any clues? Thanks.
This is an Image segmentation problem.
you can use GMM Gaussian of Mixture Model to model the image.
in your case the number of components will be 2.
after you model the image by using this mixture, you can find the probability of each pixel P(pixel x belong to the first component or the second component)
check
http://www.mathworks.com/matlabcentral/newsreader/view_thread/272162
http://www.mathworks.com/help/stats/cluster-data-from-mixture-of-gaussian-distributions.html
Currently I hope to use scale space representation to filter one image. Features in one image can be filtered using an Gaussian smooth filter with one optimal sigma. It means different features in one image can be expressed best in different scale under scale space representation.
For example, I have one image with one tree in it. In the scale space representation, three sigma values are used and they are represented as sigma0, sigma1 and sigma2. The ground is best expressed in the smoothed image with sigma0 because it contains textures mainly. The branches are best expressed in the smoother image with sigma1 and the trunk is with the smoother image with sigma2. If I hope to filter the image, I hope that the filtered pixels for the group is from the smoothed image with sigma0.
The filtered pixels for the branches are from the smoothed image with sigma1. The filtered pixels for the trunk are from the smoothed image with sigma2.
It requires that I need to determine in which smoothed image one pixel is expressed best. Is this idea plausible?
I am trying to use differece-of-Gaussian of two successive smoothed images to perform the above task. Is there any other way to combine the three smoothed image?
I use Matlab to implement the idea. The values of the three sigmas is 1.0, 2.0 and 3.0. The corresponding size of Gaussian kernel is 3, 5 and 7. I use the function fspecial to generate the kernel. Are the parameter reasonable? Please share your experience with the scale space representation to help me. You can provide some links to useful papers.
your idea is very much plausible! You are just one step away from it. I did something very similar once and it looked like this:
After smoothing your images and extracting the edges for each smoothing step (I used a weighted [to compensate for maxima supression after Gauss filtering] Sobel filter for this since DOG was not quite stable for my aplication), you can proyect (and normalize) your whole stack of edge images into a single image ("cummulative edges") which will contain the characteristic edges. You can then compare the cummulative edges image (using cross-correlation or whatever you wish) with every single image in your edge stack, the biggest value of this comparation is then the smooth-scale in which the pixel is expressed the best.
Hope that makes sense for you after reading it a couple of times.
Also don't be afraid of using much bigger kernel sizes, while it all depends on your application, I ended up using things of 51 and bigger!!! (was working with 40MP images though...)
T. Lindeberg has literally dozens of papers related to this problem. I found this one the most useful, but since you are already in the right track, I don't think reading the 50 pages will make you that much smarter. The most important part of it is maybe this one:
Principle for scale selection:
In the absence of other evidence, assume that a scale level, at which some
(possibly non-linear) combination of normalized derivatives assumes a
local maximum over scales, can be treated as reflecting a characteristic
length of a corresponding structure in the data.
I have to calculate the area, or length of the objects present in the frame.
As i use the 2d camera, the distance from the camera can't be found.
In this case, i am planning to draw a constant(X CM) line in the back ground where its length is known in CM/M.
Please find the attachment for a sample input image. (Yellow Line is a Constant line)
Consider that a person or an object stands in front of a wall, where the constant line is drawn.
Is there any way to calculate the distance of other objects with reference to the constant line?
First, it isn't a line. It is a parcel. A line is non-physical. The parcel of pixels has both area and length. The natural unit of measurement of images is pixels. Units of length are both non-physical and require assumptions.
Second, you can do a thresholded 2-d convolution. PIV-sleuth uses 2d convolution. It can allow some faster, more accurate measurement in images. Peak intensity will tell you something about the length or area. You can also use row-sum and column sum very quickly to get ideas of lengths. It helps if the images are aligned to the pixel-axes in your image. Use of affine transformations can help you test various rotations for suitability.
the output of some processing consists of a binary map with several connected areas.
The objective is, for each area, to compute and draw on the image a line crossing the area on its longest axis, but not extending further. It is very important that the line lies just inside the area, therefore ellipse fitting is not very good.
Any hint on how to do achieve this result in an efficient way?
If you have the image processing toolbox you can use regionprops which will give you several standard measures of any binary connected region. This includes
You can also get the tightest rectangular bounding box, centroid, perimeter, orientation. These will all help you in ellipse fitting.
Depending on how you would like to draw your lines, the regionprops function also returns the length for major and minor axes in 2-D connected regions and does it on a per-connected-region basis, giving you a vector of axis lengths. If you specify 4 neighbor connected you are fairly sure that the length will be exclusively within the connected region. But this is not guaranteed since `regionprops' calculates major axis length of an ellipse that has the same normalized second central moment as the connected region.
My first inclination would be to treat the pixels as 2D points and use principal components analysis. PCA will give you the major axis of each region (princomp if you have the stat toolbox).
Regarding making line segments and not lines, not knowing anything about the shape of these regions, an efficient method doesn't occur to me. Assuming the region could have any arbitrary shape, you could just trace along each line until you reach the edge of the region. Then repeat in the other direction.
I assumed you already have the binary image divided into regions. If this isn't true you could use bwlabel (if the regions aren't touching) or k-means (if they are) first.