如何掩盖图像以消除灰色噪声？

Question

I have the following raw image that I want to mask. I want just the circular shaped (almost) orange/brown structure to be masked white. How do I go about doing it?

http://imgur.com/a/HNmRn

I've tried thresholding, but I don't want the lower threshold value to be a variable.

Answer 1

You could try converting into HSV colorspace and threshold for color. But you might not be able to remove the threshold as a variable, as every image has slight variations in the lighting. From experience I can tell you that sometimes you can generously extend the threshold to fit most of the stuff you want. But a more general solution will take more sophisticated algorithms.

from opencv documentation:

11     # Convert BGR to HSV
12     hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
13 
14     # define range of blue color in HSV
15     lower_blue = np.array([110,50,50])
16     upper_blue = np.array([130,255,255])
17 
18     # Threshold the HSV image to get only blue colors
19     mask = cv2.inRange(hsv, lower_blue, upper_blue) 

For the yellowish tone you have there you will have to adjust the parameters of course.

Answer 2

Use Hough circle transform to find the the circle that separate the eye and the gray area.

The basic idea is to run Hough circle transfor and then finding the circle that has the biggest difference in values between the inside of the circles and outside.

The result:

The code:

import cv2
import numpy as np


# Read image
Irgb = cv2.imread('eye.jpg')

# Take the first channel ( No specifc reason just good contrast between inside the eye and outside)
Igray = Irgb[:,:,0]

# Run median filter to reduce noise
IgrayFilter = cv2.medianBlur(Igray,101)

# Find circles using hough circles
minRadius = np.floor(np.min(Igray.shape)/2)
circles = cv2.HoughCircles(IgrayFilter, cv2.HOUGH_GRADIENT, dp=0.5,param1=100,param2=50,minRadius=int(minRadius),minDist=100)
circles = np.uint16(np.around(circles))
cimg = Irgb

# For each circle that we found find the intinestiy values inside the circle and outside.
# We eould take the circle that as the biggest difference between inside and outside
diff = []
for i in circles[0, :]:

    # Create mask from circel identity
    mask = np.zeros_like(Igray)
    maskInverse = np.ones_like(Igray)

    cv2.circle(mask, (i[0], i[1]), i[2], 1, cv2.FILLED)
    cv2.circle(maskInverse, (i[0], i[1]), i[2], 0, cv2.FILLED)

    # Find values inside mask and outside
    insideMeanValues = np.mean(np.multiply(mask,Igray))
    outsideMeanValues = np.mean(np.multiply(maskInverse, Igray))

    # Save differnses
    diff.append(abs(insideMeanValues-outsideMeanValues))


# Take the circle with the biggest difference in color as the border circle
circleID = np.argmax(diff)
circleInfo = circles[0, circleID]

# Create mask from final image
mask = np.zeros_like(Igray)
cv2.circle(mask, (i[0], i[1]), i[2], 1, cv2.FILLED)

# Show final image only in the mask
finalImage = Irgb
finalImage[:,:,0] = np.multiply(finalImage[:,:,0],mask)
finalImage[:,:,1] = np.multiply(finalImage[:,:,1],mask)
finalImage[:,:,2] = np.multiply(finalImage[:,:,2],mask)

cv2.imwrite('circle.jpg',finalImage)