Ask Your Question

How to remove multiple hits on detection?

asked 2020-03-06 03:12:04 -0500

fabiomirisola gravatar image

updated 2020-03-10 04:27:45 -0500

Normaly houghtransfer uses mindistance between 2 circle centers but this does not work. included is my code and one of my lab results. I use multiple morphological steps to enhance the result.

My code:

# -*- coding: utf-8 -*-
Created on Mon Mar  2 14:22:03 2020

@author: Fabio Mirisola

    import numpy as np
    import cv2 as cv
    from skimage import morphology

    # Import image
    data = cv.imread("data.tif",cv.IMREAD_GRAYSCALE)

    # Import background
    background = cv.imread("background.tif",cv.IMREAD_GRAYSCALE)

    # Subtracting background from data
    cv_subt = cv.subtract(data,background)

    # Cropping the image
    Xcoord = 350 
    YCoord = 250
    Width = 1330
    Height = 1330
    crop_img = cv_subt[YCoord:YCoord+Height, Xcoord:Xcoord+Width]
    ret,thresh = cv.threshold(crop_img,25,255, cv.THRESH_BINARY)

    # Removing small pixels 
    thresh_bool = thresh.astype('bool')
    cleaned = morphology.remove_small_objects(thresh_bool, min_size=70, connectivity=5)
    im_bin = cleaned.astype('uint8')
    image = (im_bin)*255

    # Inversing the image
    ret,inverse = cv.threshold(image,5,255, cv.THRESH_BINARY_INV)

    # Opening the holes
    kernel = cv.getStructuringElement(shape=cv.MORPH_RECT, ksize=(4,4))
    opening = cv.morphologyEx(inverse, cv.MORPH_OPEN, kernel, iterations = 3)

    closing = cv.morphologyEx(opening, cv.MORPH_CLOSE, kernel, iterations = 1)
    closing_bool = closing.astype('bool')
    cleaned2 = morphology.remove_small_objects(thresh_bool, min_size=300, connectivity=400)
    cleaned2_bin = cleaned2.astype('uint8')
    image2 = (cleaned2_bin)*255

    # Detected circles

    circles = cv.HoughCircles(image2,3,cv.HOUGH_GRADIENT,1,30,param1=140,param2=45,minRadius=10,maxRadius=45)
    circles = np.uint16(np.around(circles))

    Parameters 1 and 2 don't affect accuracy as such, more reliability. 
    Param 1 will set the sensitivity; how strong the edges of the circles need to be. 
    Too high and it won't detect anything, too low and it will find too much clutter. 
    Param 2 will set how many edge points it needs to find to declare that it's found a circle. 
    Again, too high will detect nothing, too low will declare anything to be a circle. 
    The ideal value of param 2 will be related to the circumference of the circles. 

    #Drawing circles

    detected = cv.cvtColor(crop_img,cv.COLOR_GRAY2BGR)

    for i in circles[0,:]:
            #   draw    the outer   circle
            #   draw    the center  of  the circle

    #Show img
    cv.imshow("detected", detected)

    #Save image

    cv.imwrite("detected.jpg", detected)

C:\fakepath\detected.jpg In the picture you can see that it clearly detects the bubble correctly but then it detects is plenty more time.

edit retag flag offensive close merge delete


Look into non-maxima suppression or weighted fusion. You'll want to sort your results by response or score and then iterate over them, removing/suppressing or weighted fusing overlapping results that pass some overlap threshold criteria. This overlap is called IOU or intersection over union.

Der Luftmensch gravatar imageDer Luftmensch ( 2020-03-06 09:38:49 -0500 )edit

@fabiomirisola.In the picture you can see that it clearly detects the bubble correctly. The answer is NO It never detected bubbles. You can zoom out and use magnify glass to see it is not detecting bubbles. The red dots contained a lot of noises.

supra56 gravatar imagesupra56 ( 2020-03-07 08:28:42 -0500 )edit

Can you post original image?

supra56 gravatar imagesupra56 ( 2020-03-08 21:39:32 -0500 )edit

It will not accept my Jpg i try to uplaod, it does not create a fake path... the link has it

fabiomirisola gravatar imagefabiomirisola ( 2020-03-10 04:24:30 -0500 )edit

No. You can't insert into comment. You have to edit to first question and then add image.again. Thank for image.

supra56 gravatar imagesupra56 ( 2020-03-10 09:45:15 -0500 )edit

I meant original images for data.tiff and background.tiff.

supra56 gravatar imagesupra56 ( 2020-03-11 04:25:52 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2020-03-07 11:37:38 -0500

lnx gravatar image

try non-max-supression present in tensorflow like we use in YOLO algorithm

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

1 follower


Asked: 2020-03-06 03:12:04 -0500

Seen: 148 times

Last updated: Mar 10