davgit/better_color_detection_for_opencv

Easy-to-use method for color detection. This method uses multiple ranges and can automatically determine them.

Advanced color detecting tool for OpenCV

When we detect color on an image in opencv we use cv2.inRange function like in this tutorial:
https://henrydangprg.com/2016/06/26/color-detection-in-python-with-opencv/

However, sometimes accuracy of this approach is not satisfactory. Also it is usually difficult to define the range.

In this code I use multiple cv2.inRange calls in different regions of color space:

Points represent colors in {H,S,V} coordinates. Boxes represent ranges.

How it works

This code uses pre-collected data. A simple heuristic algorithm is used here to define boxes (ranges) which fit most of the points.
Basically you click on desired colors and then they are detected.

Testing

For images present here, dataset is already collected, just launch test.py to see how it works.

Collecting your own color dataset

Put your sample images to input_data folder.
Make sure that color_dataset file is empty or deleted: this code will append samples to it.
Run collect_data.py.
Images will appear on the screen. Click left mouse button on points that contain color you want to detect.
Click middle mouse button on colors which you don't want to detect. Negative points will not affect ranges. They are only needed if you want to see them on plot as showed above.

Usage

To use this code in your project you will need only color_detector.py.

from color_detector import ColorDetector

Use color dataset from file 'color_dataset' to create detector object:

cd_green = ColorDetector('color_dataset')

That's it! Now to use the detector, call get_mask:

mask = cd_green.get_mask(image)

Now you have binary mask of image.

Analysis

To see which ranges were created, write:

print(cd_green.ranges)

output is a list of two (in this example) ranges (they can be seen as boxes on the above picture):

[(array([  27.28966087,  160.95680175,   24.2726661 ]), array([  45.71033913,  226.04319825,   96.7273339 ])), (array([  27.12217638,  106.1456087 ,   24.15146678]), array([  47.87782362,  170.8543913 ,   99.84853322]))]

Alternative usage

If you already have ranges, you can use them as follows:

cd_green = ColorDetector()
cd_green.ranges = [(lower1, upper1), (lower2,upper2)] # where lowerI and upperI are numpy arrays representing colors