Now I want to extract hog feature of images, but the ratio is not the same. So I am resizing all datasets and query images into equal sizes, which is the mean size of all images and then extracting features.
I don't know if this is the right way of using HoG features, and can anyone suggest any solution for the different size of images in dataset and query image other than resizing all images? Thank you.
If you can set the parameters of the HOG descriptor in details, you can set the number of cells per image and not their size.
In this way you end up with a fixed length descriptor also for images with different size.
My database is huge. So I think setting the number of cells per image is not real.
I mean that you can set the number of cells one time, and use it for all images.
For example, dividing images in 5X10 cells, the dimensions of the cells for each image will be different, but you don't care because it is automatically computed given the image size and the cells number.
Once you have a fixed number of HOG cells, you obtain a fixed length descritpor.
Latch Shun are you calculate HOG feture to hole image or choose some object in image to create dataset?
1. Resize image by one side (width)
2.1. Crop minimum cells number in database in center of image
OR 2.2. Add borders to create standart cells number
OR 2.3. Make different classes to different sizes with quantization 7x10 -> 10x10, 6x10 -> 5x10
Hi,
First, HOG is not scale invariant. Getting the same length feature vector for each image does not guarantee the scale invariance. If you want scale invariant HOG I recommend you to read the following paper.
P. Doll´ar, S. Belongie, and P. Perona, “The fastest pedestrian detector in the west,” in Proc. British Machine Vis. Conf., 2010.
Second, it is important to have a fixed size training images with certain aspect ratio, and try to find the best HOG parameters (such as cell size, number of blocks in x, and y directions, number of orientation bins )on the object of interest (i.e. vehicles, pedestrians, etc...).
All the best,
if you have Matlab, then this may be of help
http://www.vlfeat.org/overview/hog.html
Dear Pablo
Is VFH means Viewpoint Feature Histogram?
how can we calculate total number of HOG feature vector for an image?
For example if the image size is 333 Cols and 500 rows.
Can i Extract HOG Features on a set of images containing 30 images at once?
Hi Rabelani
These 30 images are for same gesture or gesture is dynamic in nature?
They all have objects of interest nd are of the same size...gesture is dynamic
yes, all the images must of be the same size so that the HOG desriptions will the same in size.
- Badges
- Science topic