According to Lowe, David G. "Distinctive Image Features from Scale-Invariant Key Points." International Journal of Computer Vision 60.2 (2004): 91-110 "
“It is important to avoid all boundary influences in which the descriptor changes dramatically when the sample smoothly moves from one histogram to another or from one orientation to another. Therefore, trilinear interpolation is used to distribute the value of each gradient pattern to adjacent histogram cells. In other words, each basket entry is multiplied by a 1-d weight for each measurement, where d is the distance from the sample from the central bit value, measured in units of the histogram interval. "
I calculate the orientation [t] and the location of the gradient (x, y) that will be in the floating point. Currently, I'm just providing a gradient value to the values of the three-dimensional histogram [t] [x] [y] (meaning the lower bound of the floating point values t, x and y). But, according to the article, I have to distribute the gradient value to neighboring cells. I am not sure how to distribute it.
I got my answer at the following link:
HOG Trilinear interpolation of histograms
source
share