[FIXED] problem with python cv2 shape recognition code

Issue

I'm having problem with my code, it's supposed to get a two line input, the first line is how many shapes are in the image, and the second line is the image path.

The output should be the number of types of shapes, example input: there is a test1.jpg image that has 10 circles in it, the output should be: 1

my codes write here :

import cv2
import numpy as np


num_shapes = int(input("Enter the number of shapes in the image: "))
image_path = input("Enter the path to the image: ")


image = cv2.imread(image_path)

gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

ret, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY)

contours, hierarchy = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

filtered_contours = [cnt for cnt in contours if cv2.contourArea(cnt) > 100]

num_detected_shapes = len(filtered_contours)

print("Detected shapes:", num_detected_shapes)

But for this image:

The output it gives is 1.

why?

Solution

Your image has very bad contrast and the background has very high values. One way to emphasize the difference between the background and the shapes is to use OTSU thresholding instead of binary:

ret, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_OTSU)

After that, you need to invert the binary mask, since again, the highest values in your case are the background and not the shapes.

inv_binary = cv2.bitwise_not(binary)

After that, it's good to introduce some contrast by means of checking and thresholding:

kernel = 255*np.ones((10, 10), np.uint8) # dilate the binary mask, here and next line
dilated_binary = cv2.dilate(gray, kernel, iterations=1)

mask = cv2.dilate((dilated_binary < 245).astype(np.uint8)*255, kernel, iterations=1) # create another binary mask where the values are lower than 245, here I did some manual checking

contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # find the contour on the mask

filtered_contours = [cnt for cnt in contours if cv2.contourArea(cnt) > 100] # filter the contours

num_detected_shapes = len(filtered_contours)
print("Detected shapes:", num_detected_shapes)


Results -> Detected shapes: 8

Here is also an image of the contouring:

V2.0: with added shape counting

To identify the type of contour, you can approximate the shape by using cv2.approxPolyDP. The idea would be to count the vertices afterwards:

circles = 0 
triangles = 0 
rectangles = 0  
for contour in filtered_contours:
        M = cv2.moments(contour)
        if M["m00"] != 0:
            cx = int(M["m10"] / M["m00"])
            cy = int(M["m01"] / M["m00"])
        # approximate the contour
        epsilon = 0.04 * cv2.arcLength(contour, True)
        approx = cv2.approxPolyDP(contour, epsilon, True)
        # number of sides
        num_vertices = len(approx)
        # see how many sides and decide which shape
        if num_vertices == 3:
            shape = "Triangle"
            triangles += 1
        elif num_vertices == 4:
            shape = "Rectangle"
            rectangles += 1
        else: # in this case I just put the default as a circle, since there are currently no other options.
            shape = "Circle"
            circles += 1
        plt.text(cx, cy, shape)
         

After that, here is the summary:

print("Summary:") 
print("Rects or squares: "+str(rectangles)) 
print("Triangles: "+str(triangles))
print("Circles: "+str(circles))

This will give the following:

Summary:
Rects or squares: 3
Triangles: 3
Circles: 2

Answered By - Tino D