본문 바로가기

Opencv

How to use SEEDS for superpixel segmentation

반응형

** Opencv에서 제공하는 superpixel segmentation algorithm, SEEDS 사용법 **


Background: Ubuntu16.04LTS + Anaconda3 + Python3.6 + Opencv3.x

 

The algorithm SEEDS (which is originated by MV Bergh et al, 'SEEDS: Superpixels Extracted vis Energy-Driven Sampling') is a built-in function for image superpixel segmentation.

 

// CODES //////////////////////////////

import cv2
import numpy as np

# image path
dir = '/home/dooseop/PycharmProjects/Proj04/test_images/test4.bmp'


# image read
img = cv2.imread(dir) # BRG order, uint8 type
cv2.imshow('ImageWindow', img)
cv2.waitKey(0)


# convert color space
converted_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)


# set parameters for superpixel segmentation
num_superpixels = 400 # desired number of superpixels
num_iterations = 4 # number of pixel level iterations. The higher, the better quality
prior = 2 # for shape smoothing term. must be [0, 5]
num_levels = 4
num_histogram_bins = 5 # number of histogram bins
height, width, channels = converted_img.shape


# initialize SEEDS algorithm
seeds = cv2.ximgproc.createSuperpixelSEEDS(width, height, channels, num_superpixels, num_levels, prior, num_histogram_bins)


# run SEEDS
seeds.iterate(converted_img, num_iterations)


# get number of superpixel
num_of_superpixels_result = seeds.getNumberOfSuperpixels()
print('Final number of superpixels: %d' % num_of_superpixels_result)


# retrieve the segmentation result
labels = seeds.getLabels() # height x width matrix. Each component indicates the superpixel index of the corresponding pixel position


# draw contour
mask = seeds.getLabelContourMask(False)
cv2.imshow('MaskWindow', mask)
cv2.waitKey(0)


# draw color coded image
color_img = np.zeros((height, width, 3), np.uint8)
color_img[:] = (0, 0, 255)
mask_inv = cv2.bitwise_not(mask)
result_bg = cv2.bitwise_and(img, img, mask=mask_inv)
result_fg = cv2.bitwise_and(color_img, color_img, mask=mask)
result = cv2.add(result_bg, result_fg)
cv2.imshow('ColorCodedWindow', result)
cv2.waitKey(0)


cv2.destroyAllWindows()

The final results are