import numpy as np
import cv2 as cv
from sklearn.cluster import MeanShift, estimate_bandwidth
img = cv.imread(your_image)
# filter to reduce noise
img = cv.medianBlur(img, 3)
# flatten the image
flat_image = img.reshape((-1,3))
flat_image = np.float32(flat_image)
# meanshift
bandwidth = estimate_bandwidth(flat_image, quantile=.06, n_samples=3000)
ms = MeanShift(bandwidth, max_iter=800, bin_seeding=True)
ms.fit(flat_image)
labeled=ms.labels_
# get number of segments
segments = np.unique(labeled)
print('Number of segments: ', segments.shape[0])
# get the average color of each segment
total = np.zeros((segments.shape[0], 3), dtype=float)
count = np.zeros(total.shape, dtype=float)
for i, label in enumerate(labeled):
total[label] = total[label] + flat_image[i]
count[label] += 1
avg = total/count
avg = np.uint8(avg)
# cast the labeled image into the corresponding average color
res = avg[labeled]
result = res.reshape((img.shape))
# show the result
cv.imshow('result',result)
cv.waitKey(0)
cv.destroyAllWindows()
下面是来自sklearn的一个结果:
请注意,首先平滑图像以减少噪波。此外,这与图像分割论文中的算法不完全相同,因为图像和核是平坦的
代码如下:
相关问题 更多 >
编程相关推荐