Recently, I do some RGB images processing. Using the gradients of these images to calculate the HOG features of the images, however, I am confused about how to calculate the gradient of color images, do not tell me about transfer to grayscale image first because I do not want to use that kind of method to calculate the gradient. I want to use one vector to represent the gradient of the color images, is it possible? Please give me some link or references to this? Thank you very much
If you work in the matlab, you can apply gradient operator independently on each one of the colors. And in the following way obtain gradient map for each one of the colors:
[imRX imRY]=gradient(pic(:,:,1));
[imGX imGY]=gradient(pic(:,:,2));
[imBX imBY]=gradient(pic(:,:,3));
Good luck!
Generally speaking, a color image can be considered as a vector function that associates to each point in R^2 (x1,x2) - pixel - a point in R^3 - color, i.e. (x_1, x_2) -> (f_1(x_1,x_2), f_2(x_1,x_2), f_3(x_1,x_2)) . The gradient is therefore the Jacobian matrix, i.e. the3x2 matrix with enties the partial derivative (df_i / dx_j), i=1,2,3, j=1,2.
Hi,
please check the following link
http://link.springer.com/chapter/10.1007%2F3-540-63507-6_190#page-1
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1352955&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D1352955
http://sibgrapi.sid.inpe.br/col/sid.inpe.br/banon/2002/11.13.12.22/doc/316-323.pdf
ftp://ftp.math.ucla.edu/pub/camreport/cam08-40.pdf
You can have look at the following link for DPM MATLAB source code which uses HOG features based on RGB input image.
http://www.cs.berkeley.edu/~rbg/latent/
or as mentioned in the following HOG extractor
http://bwasti.com/418
https://github.com/mofarrell/HOGextractor
They always call it as Pedro's HOG features "referring to author: Pedro Felzenswalb" in DPM ref: P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan, "Object Detection with Discriminatively Trained Part Based Models," in PAMI, Sep. 2010
For RGB image, you can calculate Pedro's HOG features by calculating the gradient in each of three channels then you take the values in the channel with largest magnitude + some other processing.
https://github.com/mofarrell/HOGextractor/blob/master/features_pedro.cc
Look at this link.
http://homepages.inf.ed.ac.uk/rbf/HIPR2/sobel.htm
I go with Ali. Histogram of Gradient (HoG) is the right choice.
Hi Forrest,
You can use Mathematical Morphology: for example in Matlab:
I=imread('image.bmp');
se = strel('disk',1);
erod = imerode(I,se);
dilat=imdilate(I,se);
gradient=dilat-erod;
or
gradient=I-erod;
or
gradient=I-dilat;
You can inspire you from the two codes in attached files
According to the HOG paper you can compute gradients in all color channels, pick the highest one.
See this link:
http://vision.stanford.edu/teaching/cs231b_spring1213/slides/hog_rafael_tao.pdf
The common way is to calculate the gradient of ech plane (R, G and B). Then, you can calculate the magnitide of all.
- Badges
- Science topic
- Similar topics
- Vectorization