HOG-BGR-with-visualization

HOG feature descriptor, a kind of feature transform before we put our image into SVM or NN.

In this API, I split hog feature into BGR channel respectively, it's a little different from hog in skimage, but this API still support grayscale image.
This API divides orients of gradient into 9 bins and each bin represent 0 20 40 60 ... 160.

Furthermore, due to hog's performance has some relationship with tuning parameters, so this repository provides hog visualization both before and after doing block normalization.

Environment

Python>=3.9.1
numpy>=1.20.0
opencv-python>=4.5.1.48

Installation

pip install hogBGR  

Demo

Original Picture

import cv2
from hogBGR import HOG
from hogBGR import HOGvisualized
from hogBGR import util

HOG

Let's create a 16x16 (pixels) cell, 2x2 (cells) block hog image.

img = cv2.imread(r'hogBGR/example/rose.png')
cell_hogfeature = HOG.Cell_HOG(img,cell_size=(16,16))
hog = HOG.BlockNorm_HOG(cell_hogfeature,block_size=(2,2))

or

img = cv2.imread(r'hogBGR/example/rose.png')
hog = HOG.HOGfeature(img,cell_size=(16,16),block_size=(2,2),flatten=False)

Take a visualization

res = HOGvisualized.visualize_HOG(hog[:,:,:,0,0,:],15)
cv2.imshow("HOG feature with block normaliztion",res)

or

util.BlockNormaliztionHOG_Show(img,one_cell=True)

Below shows hog image in each cell of blocks. (First method would only show first cell of blocks)

HOG without block normalization

Let's create a 16x16 (pixels) cell hog image.

img = cv2.imread(r'hogBGR/example/rose.png')
cell_hogfeature = HOG.Cell_HOG(img,cell_size=(16,16))

Take a visualization

res = HOGvisualized.visualize_HOG(cell_hogfeature,15)
cv2.imshow("part1 - HOG feature without block normaliztion",res)

or

util.NonBlockNormaliztionHOG_Show(img)

Below shows hog image before block normalization

Gray scale input image is also supported

Orginal Picture

Visualization

Simplified visualization

You can try the result with yourself if you want a faster visualization.

res = HOGvisualized.Fast_visualize_HOG(hogfeature,7)