Histogram of Luv image

I need to compute a single histogram with 125 bins (5^3) for an Luv image. I can get separate channels using L=img(:,:,1) and so on and get a 1D array against them as L=L(:). And i can use histc function with setting r = [5 5 5] and passing 0:prod(r)-1 as second argument (binranges). But I don't understand how the L,u and v channels should be consolidated into a single vector to pass as first argument to histc so i get right histogram. Some help please.

Answer 1

[Image Analyst]

Simply use histogram:

histObject = histogram(img);

It will treat img and one single object - all the L, U, and V channels lumped together with no distinction. I have no idea why you'd want to do this deceptive thing however.

Or you can use histcounts(). histc() is deprecated so you should use these two new functions.

[Maria Kanwal]

Thank you Image Analyst. But why did you call it "deceptive thing"?

[Image Analyst]

Because L has different meaning and units and range than u and v. It would be like if you had measurements of individuals heights, years of birth, and heart rates, and you said "I don't want histograms of each of them, let's just lump all those numbers together and take a histogram of the whole group." See - it's meaningless and deceptive.

[Maria Kanwal]

Right, here is the problem statement in brief

i have video frames each segmented into certain objects, i need to track all objects starting from their first appearance so i got properties of all objects in frame 1 using regionprops(), from here i stored centroid(1) and (2) and cropped each object from image using corresponding boundingbox (got histogram for this sub image). Now 2nd frame also goes through same process and i have centroid(1), (2) and its histogram. for comparison of each object of frame 2 with all objects in frame 1: first i used euclidean distance between centroids:

dist=sqrt((frame1.Centroid(1)-frame2.Centroid(1)).^2+(frame1.Centroid(2)-frame2.Centroid(2)).^2);

if dist<= distance threshold, objects possibly match so we call all objects of frame 1 within this distance to be candidate matches. then i want to use cosine value of angle between histograms of objects, that's where i want to compute histogram. the code i posted before computed histogram for hsv image (using hsv doesn't give good tracking results and the dataset i'll be using this on is DIEM , it has all sorts of videos so i am not giving any images) but that does not work for Luv (because of different meanings of these colorspace i think). can you kindly suggest computing Luv image histogram which can retain meaning?

[Image Analyst]

There are many features that you can measure from each blob. It doesn't have to be just 2 (centroid and histogram). You may need to measure lots of things and come up with some sort of weighting or voting scheme to correspond each blob in a frame with blobs from prior frames. Have you tried the tracking methods available in the Computer Vision System Toolbox? https://www.mathworks.com/products/computer-vision/features.html#object-tracking-and-motion-estimation

[Maria Kanwal]

I had tried these but nothing worked close to good enough (i suppose because of diversity of objects: ball, bird, human, violon stick etc.), mostly feature detection and extraction methods come out empty (i mean there weren't much points to be matched). one thing i found to be best was 'template matching', problem with that approach is how to check for new objects entering the frame, also it finds objects which are actually not in the image(bigger problem). before histograms i also tried euclidean distance of centroid and mean values of 3 channels (Lab or Luv color space), that gave poor results than histograms.

[Image Analyst]

Yep, you're starting to realize all the problems that can come into effect once you start working on real world situations. Now, just imagine if your camera was also moving, like it was attached to a car. So then virtually everything in your image is moving. Yes, it can get quite complicated. Sorry I can't help you more. Perhaps you can hire a university or think tank (like SRI International to develop a system for you.

[Image Analyst]

There was no sarcasm there. Sorry you misinterpreted it. It's not unusual for people to hire experts to carry out projects for a variety of reasons - it's done all the time. We do it extensively in my company.

[Maria Kanwal]

oh i am sorry about that.