[英]How to crop or remove white background from an image
I am trying to compare images using OpenCV and Python.我正在尝试使用 OpenCV 和 Python 比较图像。
Consider these images:考虑这些图像:
Both feature an identical pair of shoes, set to a white background.两者都有一双相同的鞋子,设置为白色背景。 The only difference being that the first has a taller background than the second.
唯一的区别是第一个的背景比第二个高。
I want to know how to programmatically crop the white backgrounds of both so that I'm left with only the pair of shoes.我想知道如何以编程方式裁剪两者的白色背景,这样我就只剩下那双鞋了。
I must add that it won't be possible for me to manually crop the backgrounds.我必须补充一点,我无法手动裁剪背景。
You requirement in the comment: The shoes are on a white background. I would like to completely get rid of the border; as in be left with a rectangular box with either a white or a transparent background, having the length and width of the shoes in the picture.
您在评论中要求:
The shoes are on a white background. I would like to completely get rid of the border; as in be left with a rectangular box with either a white or a transparent background, having the length and width of the shoes in the picture.
The shoes are on a white background. I would like to completely get rid of the border; as in be left with a rectangular box with either a white or a transparent background, having the length and width of the shoes in the picture.
Then my steps to crop the target regions:然后我裁剪目标区域的步骤:
- Convert to gray, and threshold
转换为灰度和阈值
- Morph-op to remove noise
Morph-op 消除噪音
- Find the max-area contour
找到最大面积轮廓
- Crop and save it
裁剪并保存
#!/usr/bin/python3
# Created by Silencer @ Stackoverflow
# 2018.01.23 14:41:42 CST
# 2018.01.23 18:17:42 CST
import cv2
import numpy as np
## (1) Convert to gray, and threshold
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
th, threshed = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY_INV)
## (2) Morph-op to remove noise
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (11,11))
morphed = cv2.morphologyEx(threshed, cv2.MORPH_CLOSE, kernel)
## (3) Find the max-area contour
cnts = cv2.findContours(morphed, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[-2]
cnt = sorted(cnts, key=cv2.contourArea)[-1]
## (4) Crop and save it
x,y,w,h = cv2.boundingRect(cnt)
dst = img[y:y+h, x:x+w]
cv2.imwrite("001.png", dst)
Result:结果:
Kinght's solution works well. Kinght 的解决方案运行良好。 In my case, I also have CMYK images.
就我而言,我也有 CMYK 图像。 When I crop them, I get incorrect (vivid colors) output.
当我裁剪它们时,我得到了不正确的(鲜艳的颜色)输出。 And it seems OpenCV doesn't support CMYK.
而且似乎 OpenCV 不支持 CMYK。 So I needed a way to convert CMYK images to RGB, and then open it with OpenCV.
所以我需要一种方法将 CMYK 图像转换为 RGB,然后用 OpenCV 打开它。 This is how I handled it:
我是这样处理的:
import cv2
import numpy
from PIL import Image
from PIL import ImageCms
# force opening truncated/corrupt image files
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
img = "shoes.jpg"
img = Image.open(img)
if img.mode == "CMYK":
# color profiles can be found at C:\Program Files (x86)\Common Files\Adobe\Color\Profiles\Recommended
img = ImageCms.profileToProfile(img, "USWebCoatedSWOP.icc", "sRGB_Color_Space_Profile.icm", outputMode="RGB")
# PIL image -> OpenCV image; see https://stackoverflow.com/q/14134892/2202732
img = cv2.cvtColor(numpy.array(img), cv2.COLOR_RGB2BGR)
## (1) Convert to gray, and threshold
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
th, threshed = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY_INV)
## (2) Morph-op to remove noise
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (11,11))
morphed = cv2.morphologyEx(threshed, cv2.MORPH_CLOSE, kernel)
## (3) Find the max-area contour
cnts = cv2.findContours(morphed, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[-2]
cnt = sorted(cnts, key=cv2.contourArea)[-1]
## (4) Crop and save it
x,y,w,h = cv2.boundingRect(cnt)
dst = img[y:y+h, x:x+w]
# add border/padding around the cropped image
# dst = cv2.copyMakeBorder(dst, 10, 10, 10, 10, cv2.BORDER_CONSTANT, value=[255,255,255])
cv2.imshow("image", dst)
cv2.waitKey(0)
cv2.destroyAllWindows()
# create/write to file
# cv2.imwrite("001.png", dst)
by PIL you can convert white background to transparent:通过 PIL,您可以将白色背景转换为透明:
from PIL import Image
def convertImage():
img = Image.open("hi.png")
img = img.convert("RGBA")
datas = img.getdata()
newData = []
for item in datas:
if item[0] == 255 and item[1] == 255 and item[2] == 255:
newData.append((255, 255, 255, 0))
else:
newData.append(item)
img.putdata(newData)
img.save("./New.png", "PNG")
print("Successful")
convertImage()
This link worked perfectly for me for a similar problem, although it uses PIL.这个链接对我来说非常适合解决类似的问题,尽管它使用了 PIL。 Note that it will result in a rectangular image, bounded by the top/right/bottom/left-most pixels that are not white.
请注意,这将产生一个矩形图像,由非白色的顶部/右侧/底部/最左侧像素界定。 In your case, it should give identical images with the same size.
在您的情况下,它应该提供相同大小的相同图像。
I am guessing the code could be easily adapted to work with OpenCV functions only.我猜代码可以很容易地调整为仅适用于 OpenCV 函数。
I found this on github.我在github上找到了这个。
https://imagemagick.org/script/download.php https://imagemagick.org/script/download.php
import pgmagick
def remove_background(image, background=None):
"""Returns a copy of `image` that only contains the parts that is distinct
from the background. If background is None, returns parts that are
distinct from white."""
if background is None:
background = pgmagick.Image(image.size(), 'white')
elif isinstance(background, pgmagick.Image):
blob = pgmagick.Blob()
background.write(blob)
background = pgmagick.Image(blob, image.size())
else:
background = pgmagick.Image(image.size(), background)
background.composite(image, 0, 0, pgmagick.CompositeOperator.DifferenceCompositeOp)
background.threshold(25)
blob = pgmagick.Blob()
image.write(blob)
image = pgmagick.Image(blob, image.size())
image.composite(background, 0, 0, pgmagick.CompositeOperator.CopyOpacityCompositeOp)
return image
Although the question has been answered thoroughly already, I would like to share a simple version that relies on numpy only:虽然这个问题已经得到了彻底的回答,但我想分享一个仅依赖于 numpy 的简单版本:
import numpy as np
def remove_background(image, bg_color=255):
# assumes rgb image (w, h, c)
intensity_img = np.mean(image, axis=2)
# identify indices of non-background rows and columns, then look for min/max indices
non_bg_rows = np.nonzero(np.mean(intensity_img, axis=1) != bg_color)
non_bg_cols = np.nonzero(np.mean(intensity_img, axis=0) != bg_color)
r1, r2 = np.min(non_bg_rows), np.max(non_bg_rows)
c1, c2 = np.min(non_bg_cols), np.max(non_bg_cols)
# return cropped image
return image[r1:r2+1, c1:c2+1, :]
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.