def detector(src, size, sigma, threshold1, threshold2):
#Step 1: Filter out noise by Gaussian filtering
g_filter = ip.getGaussianKernel(size, sigma)
img_smoothed = ip.convolve(src, g_filter)
#Step 2: Find gradient intensity and directions:
gradient_intensity, gradient_direction = calculateGradient(img_smoothed)
#Step 3: Non-max Suppresion (thinning):
nonmax_suppressed = nonmax_suppr(gradient_intensity, gradient_direction)
#Step 4 and 5: Double Thresholding and Edge Tracking
lnkd, thresholded = hysteresis(nonmax_suppressed, gradient_direction, threshold1, threshold2)
return thresholded
def calculateGradient(image):
operator_y, operator_x = ip.getGradientOperator(ip.SOBEL)
intensity_y = ip.convolve(image, operator_y, True)
intensity_x = ip.convolve(image, operator_x, True)
#I_xy=sqrt(I_x^2+I_y^2) can be approximated as I_xy=abs(I_x)+abs(I_y)
#approximation: gradient_strength = np.abs(intensity_y) + np.abs(intensity_x)
gradient_intensity = np.hypot(intensity_y, intensity_x)
#gradient_intensity = np.abs(intensity_y)+np.abs(intensity_x)
gradient_direction = np.arctan2(intensity_y, intensity_x)
#round directions to 0, 45, 90, 135 degrees DIRECTION QUANTIZATION
gradient_direction = (np.round(gradient_direction*180/np.pi/45)*45)
for y in range(gradient_direction.shape[0]):
for x in range(gradient_direction.shape[1]):
if gradient_direction[y,x] == 180:
gradient_direction[y,x]=0
if gradient_direction[y,x] < 0:
gradient_direction[y,x]+=180
return gradient_intensity, gradient_direction
img1 = imread('data/I1.jpg')
img2 = imread('data/I2.jpg')
#convert to grayscale, color features are irrelevant for edge detection
img1 = ip.color2gray(img1)
img2 = ip.color2gray(img2)
img2x = np.copy(img2)
img2y = np.copy(img2)
#get operators
roberts_y, roberts_x = ip.getGradientOperator(ip.ROBERTS)
prewitt_y, prewitt_x = ip.getGradientOperator(ip.PREWITT)
sobel_y, sobel_x = ip.getGradientOperator(ip.SOBEL)
#convolve with operators
img1_roberts_x = ip.convolve(img1, roberts_x, no_abs=True)
img1_roberts_y = ip.convolve(img1, roberts_y, no_abs=True)
img1_prewitt_x = ip.convolve(img1, prewitt_x, no_abs=True)
img1_prewitt_y = ip.convolve(img1, prewitt_y, no_abs=True)
img1_sobel_x = ip.convolve(img1, sobel_x, no_abs=True)
img1_sobel_y = ip.convolve(img1, sobel_y, no_abs=True)
#get gradient strength and directions
img1_roberts_Gs, img1_roberts_Gd = ip.getGsGd(
ip.convolve(img1, roberts_y, no_abs=True),
ip.convolve(img1, roberts_x, no_abs=True), ip.ROBERTS)
img1_prewitt_Gs, img1_prewitt_Gd = ip.getGsGd(img1_prewitt_y,
img1_prewitt_x, ip.PREWITT)
img1_sobel_Gs, img1_sobel_Gd = ip.getGsGd(img1_sobel_y, img1_sobel_x,
ip.SOBEL)