Python - OpenCV connectedComponents - user input + selecting feature/label

Question

I have two images, let's call them image 1 and image 2. I can use the following to select a random feature in image 1 and display it:

segall_prev = cv2.imread(str(Path(seg_seg_output_path, prev_filename)), 0);
segall_prev = np.uint8(segall_prev)
ret, thresh = cv2.threshold(segall_prev, 127, 255, 0)
numLabels, labelImage, stats, centroids = cv2.connectedComponentsWithStats(thresh, 8)

# Pick random prev-seg feature
random_prev_seg = random.randint(0, np.amax(labelImage))

i = np.unique(labelImage)[random_prev_seg]
pixels = np.argwhere(labelImage == random_prev_seg)
labelMask = np.zeros(thresh.shape, dtype="uint8")
labelMask[labelImage == i] = 1
numPixels = cv2.countNonZero(labelMask)

# Display chosen feature from prev_seg image only
fig_segall_prev = plt.figure()
fig_segall_prev_ = plt.imshow(labelMask)
plt.title(prev_filename + ' prev_seg')

Which will display an image such as:

So the idea is that that is the feature from the previous frame (image 1), and then the user will select where that feature is in the next frame (image 2) - basically tracing the object across frames.

# Display seg image and allow click
seg_ret, seg_thresh = cv2.threshold(segall, 127, 255, 0)
seg_numLabels, seg_labelImage, seg_stats, seg_centroids = cv2.connectedComponentsWithStats(seg_thresh, 8)
        
mutable_object = {} 

def onclick(event):
   # Capture click pixel location
   X_coordinate = int(event.xdata)
   Y_coordinate = int(event.ydata)
   mutable_object['click'] = X_coordinate
   print('x= ' + str(X_coordinate))
   print('y= ' + str(Y_coordinate))
            
   # Compare captured location with feature locations
   x = np.where(seg_labelImage == 1)
            
   #print(x[0])
   if X_coordinate in x[0] and Y_coordinate in x[1]:
      print('yes')
                
fig_segall = plt.figure()
cid = fig_segall.canvas.mpl_connect('button_press_event', onclick)
fig_segall_ = plt.imshow(seg_labelImage)
plt.title(filename + ' seg')

When image 2 is shown it is as follows:

So my question is, how do I go about capturing the location where the user clicks in the second image and then check whether those x-y coordinates correspond to where a feature is as found by cv2.connectedComponentsWithStats? I then need to save just that feature as a separate image and then use the location of the selected feature for use in the corresponding colour image.

If anyone can help, or suggest ways to improve the code - because it is very messy as I've been trying to figure this out for hours now... then that would be much appreciated. Thanks!

The solution (thanks to asdf):

        seg_numLabels, seg_labelImage, seg_stats, seg_centroids = cv2.connectedComponentsWithStats(seg_thresh, 8)
        
        selected_features = {}
        def onclick(event):
            # Capture click pixel location
            X_coordinate = int(event.xdata)
            Y_coordinate = int(event.ydata)
            
            print(f'{X_coordinate=}, {Y_coordinate=}')       
        
            obj_at_click = seg_labelImage[Y_coordinate, X_coordinate]  # component label at x/y, 0 is background, above are components
            if obj_at_click != 0:  # not background
                # clicked on a feature
                selected_feature = np.unique(seg_labelImage) == obj_at_click  # get only the selected feature
                selected_features[filename] = selected_feature
                print(f'Saved feature number {obj_at_click}')
            else:
                print('Background clicked')
            
        # Display seg image
        fig_segall = plt.figure()
        fig_segall.canvas.mpl_connect('button_press_event', onclick)
        plt.imshow(segall)
        plt.title(filename + ' seg')

With no detailed information such as camera model or motion of camera or object, it will be hard or impossible to predict where the object will be. You might have success by get the shape and trying to match shapes. Connected components has shape metrics that can be exported if you use the skimage region props at https://scikit-image.org/docs/dev/api/skimage.measure.html#skimage.measure.regionprops — fmw42, Jun 14 '22 at 23:37

asdf · Accepted Answer · 2022-06-15T16:45:57.063

cv2.connectedComponentsWithStats labels all connected components of a binary image with numbers from 1 to n_components (0 is background). So you just need to check if the x and y coordinates of the mouse click lie within a segmented component. The following code should demonstrate how to get a binary mask of the component that was clicked:

selected_features = {}
img_name = 'example1'

def onclick(event):
    # Capture click pixel location
    X_coordinate = int(event.xdata)
    Y_coordinate = int(event.ydata)
    
    print(f'{X_coordinate=}, {Y_coordinate=}')       

    obj_at_click = seg_labelImage[y, x]  # component label at x/y, 0 is background, above are components
    if obj_at_click != 0:  # not background
        # clicked on a feature
        selected_feature = seg_labelImage == obj_at_click  # get only the selected feature
        selected_features[img_name] = selected_feature
        print(f'Saved feature number {obj_at_click}')
    else:
        print('Background clicked')

I can't test the code and I've never worked with matplotlib events, so I could be that the returned click coordinates need to be offset or scaled to get the correct image coordinates. You might want to display the image at its original size and without ticklabels etc. This stackoverflow answer might help you with this.

Let me know if I didn't get your question right!

That solved it ! :D Thanks. I had to invert the X and Y in the obj_at_click = seg_labelImage[x, y] bit so it reads [y, x] instead. I've included the finished code in my answer. — Connor, Jun 15 '22 at 10:22

Python - OpenCV connectedComponents - user input + selecting feature/label

1 Answers1