Color Signature Tool

Overview

Small Python GUI for extracting HSV values from an image. Run the script, load an image, select pixels. HSV values distributions will appear in a violin chart. Choose the best min/max values of each channel for building your filter.

HSV color channel values have the following ranges, which are the same as in OpenCV. [1]

H -> 0-179
S -> 0-255
V -> 0-255

Dependencies

Python 3
matplotlib
numpy
opencv

Install & Tutorial

1 - Clone this repository

git clone https://github.com/martin0004/color_signature_tool.git
cd color_signature_tool

2 - Install packages. Make sure to install Anaconda first

conda create --name color-signature-tool --channel conda-forge python=3.9 numpy matplotlib opencv
conda activate color-signature-tool

3 - Run the tool.

python color_signature_tool.py

4 - A small GUI will open. Click on Import File and browse for your image.

5 - Click on pixels for which you want to know HSV values. As you click on pixels, the violin chart on the right will populate. Larger sections of a violin indicate more pixels with this value. In this example I am clicking on the duckies.

6 - If you have a 2nd images of the same object (say in different lighting conditions), import it and keep clicking on pixels. HSV values will be added to the ones which were already extracted.

7 - Then use engineering judgment to decide which range of values you want to use in your filter. In the example above, most of the V channnel values seem to be between 210 and 255. Values below 210 are most likely noise (or maybe one of the pixels I clicked on was just outside a duckie, so from a different color). Therefore I only use range 210-255 for channel V in my filter.

8 - Finally, build a filter with your favorite image library. The example below uses the Python cv2 module.

$ cat filter_example.py

import cv2
import matplotlib.pyplot as plt
import numpy as np

# BGR and HSV images
bgr = cv2.imread("../images/duckies_and_tomatoes.jpg")
hsv = cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)
# Matplotlib needs RGB image
rgb = cv2.cvtColor(bgr, cv2.COLOR_BGR2RGB)

h_min = 15
h_max = 35
s_min = 65
s_max = 180
v_min = 210
v_max = 255

lower = np.array([h_min, s_min, v_min])
upper = np.array([h_max, s_max, v_max])

mask = cv2.inRange(hsv, lower, upper)

rgb_masked = np.copy(rgb)
rgb_masked[mask == 0] = [0,0,0]

plt.imshow(rgb_masked)
plt.show()

$ python3 filter_example.py

Possible improvements

• Allow the user to select from more color channels and ranges from a picklist:

  RGB [255/255/255] -> RGB with channel values from 0-255
  BGR [255/255/255] -> BGR with channel values form 0-255
  RGBA [...] -> ...
  HSV [179/255/255]
  HSV [360/100/100]
  ...

• Allow the user to create different sets of HSV signatures. As an example, the user could click on the duckies, then click on button "New set", then on the tomatoes, and the HSV values for tomatoes would be a different set of violins on the chart. This would help see overlap of signatures between different objects.

References

[1] Educba, Introduction to OpenCV HSV range, https://www.educba.com/opencv-hsv-range/