如何在python中计算图像上的红色和黄色多边形？

Question

I have images where some elements are boxed with either red or yellow. The box is a polygon in general.

The objective is to count the number of elements which are red and yellow.

How can I do this count ?

I have tried to apply a mask that gets the red color or the yellow color but it gave relatively poor results. But I have no idea how to do the count later.

mask_example

Answer 1

You will probably be looking at steps like the following:

• Step 1 - discriminate red from yellow
• Step 2 - mask out red and do remaining steps, then mask out yellow and do remaining steps
• Step 3 - fill shape outlines to make complete and continuous
• Step 4 - find contours of shapes and count them

There are many ways of doing each step, and I have not tried them all. As I have numbered them, other folk may wish to make suggestions for improving any step.

---

So, for Step 1, you could discriminate red from yellow by calculating each pixel in a new image as:

new = (Red - Green)/(Red + Green + 1)

The +1 is just to stop division by zero. I actually did that with ImageMagick like this, but you could do it with PIL/Pillow + Numpy, or OpenCV just the same:

convert boxes.png -fx "(u.r-u.g)/(u.r+u.g+1)" -auto-level result1.png

Hopefully you can see that makes the yellow pixels darker and the red ones lighter.

That will then give you a threshold that you can use to turn on and off each colour separately. The a channel in Lab colourspace is also a reasonable discriminant, by the way.

---

Step 2 is just a thresholding and masking operation since we can see the separating threshold from Step 1.

---

Step 3 means filling out the boxes. You could do that with "morphology" , so I could close the gaps in the white horizontal lines with a horizontal 15x1 pixel structuring element like this:

convert result1.png -threshold 50%  -morphology close rectangle:15x1  result2.png

And then with a vertical 1x15 structuring element as well like this:

convert result1.png -threshold 50%  -morphology close rectangle:15x1 -morphology close rectangle:1x15 result2.png

to get this, where you can see the gaps are all closed and the outline is complete:

Both OpenCV and scikit-image have morphology functions you could use for this.

---

Step 4 - you can now use OpenCV's findContours() to count the blobs and get their characteristics.