使用语义分割进行图像前景后景分离处理

在我们以前的文章中,我们学习了什么是语义分割,以及如何在PyTorch中使用DeepLabv3来获得图像中检测到的对象的RGB掩码。即我们将图像中的对象标识出来。

虽然语义分割是很酷,但让我们看看如何在一些现实世界的应用程序中使用语义分割的处理结果。 在这篇文章中,我们将在torchvision中使用DeepLabv3制作以下应用程序。

1、去掉背景

2、改变背景

3、模糊背景

4、背景灰度化

如果您没有阅读我们以前关于使用torchvision进行语义分割的帖子,请最好先阅读下,因为我们将修改该帖子中使用的一些方法,并跳过该帖子中包含的一些详细注解信息。

1、删除背景

让我们选择一个真实的应用程序进行演示。 假设你拥有一个在线销售二手车的网站。. 你的网站上的卖家上传他们的二手车的照片。由于大多数卖家都不是老练的摄影师,所以你不能期望他们拍出有良好背景的照片。您可能还希望你的网页风格一致,要求所有的图片有相同的纯色背景。 在这个例子中,我们将看到如何删除汽车图片上的背景,在下面的部分中,我们将使用DeepLabV3来实现这一点。先看完整代码:

import os
from google.colab import drive
drive.mount('/content/drive')
path = "/content/drive/My Drive/Colab Notebooks"
os.chdir(path)
os.listdir(path)

from torchvision import models
from PIL import Image
import matplotlib.pyplot as plt
import torch
import numpy as np
import cv2

# Apply the transformations needed
import torchvision.transforms as T

# Define the helper function
def decode_segmap(image, source, nc=21):
    label_colors = np.array([(0, 0, 0),  # 0=background
                             # 1=aeroplane, 2=bicycle, 3=bird, 4=boat, 5=bottle
                             (128, 0, 0), (0, 128, 0), (128, 128, 0), (0, 0, 128), (128, 0, 128),
                             # 6=bus, 7=car, 8=cat, 9=chair, 10=cow
                             (0, 128, 128), (128, 128, 128), (64, 0, 0), (192, 0, 0), (64, 128, 0),
                             # 11=dining table, 12=dog, 13=horse, 14=motorbike, 15=person
                             (192, 128, 0), (64, 0, 128), (192, 0, 128), (64, 128, 128), (192, 128, 128),
                             # 16=potted plant, 17=sheep, 18=sofa, 19=train, 20=tv/monitor
                             (0, 64, 0), (128, 64, 0), (0, 192, 0), (128, 192, 0), (0, 64, 128)])
    r = np.zeros_like(image).astype(np.uint8)
    g = np.zeros_like(image).astype(np.uint8)
    b = np.zeros_like(image).astype(np.uint8)
    for l in range(0, nc):
        idx = image == l
        r[idx] = label_colors[l, 0]
        g[idx] = label_colors[l, 1]
        b[idx] = label_colors[l, 2]
    rgb = np.stack([r, g, b], axis=2)
    # Load the foreground input image
    foreground = cv2.imread(source)
    print('foreground1')
    print(foreground)
    # Change the color of foreground image to RGB
    # and resize image to match shape of R-band in RGB output map
    foreground = cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB)
    foreground = cv2.resize(foreground, (r.shape[1], r.shape[0]))
    print('foreground2')
    # Create a background array to hold white pixels
    # with the same size as RGB output map
    background = 255 * np.ones_like(rgb).astype(np.uint8)
    # Convert uint8 to float
    foreground = foreground.astype(float)
    background = background.astype(float)
    # Create a binary mask of the RGB output map using the threshold value 0
    th, alpha = cv2.threshold(np.array(rgb), 0, 255, cv2.THRESH_BINARY)
    # Apply a slight blur to the mask to soften edges
    print('alpha1')
    print(alpha)
    alpha = cv2.GaussianBlur(alpha, (7, 7), 0)
    print('alpha2')
    print(alpha)
    # Normalize the alpha mask to keep intensity between 0 and 1
    alpha = alpha.astype(float) / 255
    print('alpha3')
    print(alpha)
    # Multiply the foreground with the alpha matte
    foreground = cv2.multiply(alpha, foreground)
    # Multiply the background with ( 1 - alpha )
    background = cv2.multiply(1.0 - alpha, background)
    # Add the masked foreground and background
    outImage = cv2.add(foreground, background)
    # Return a normalized output image for display
    # return outImage.astype(int)
    return outImage / 255


def segment(net, path, show_orig=True, dev='cuda'):
    img = Image.open(path)
    if show_orig:
        plt.imshow(img)
        plt.axis('off')
        plt.show()
    # Comment the Resize and CenterCrop for better inference results
    trf = T.Compose([T.Resize(450),
                     # T.CenterCrop(224),
                     T.ToTensor(),
                     T.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])])
    inp = trf(img).unsqueeze(0).to(dev)
    out = net.to(dev)(inp)['out']
    om = torch.argmax(out.squeeze(), dim=0).detach().cpu().numpy()
    rgb = decode_segmap(om, path)
    plt.imshow(rgb)
    plt.axis('off')
    plt.show()

dlab = models.segmentation.deeplabv3_resnet101(pretrained=1).eval()
segment(dlab, 'bluecar.png', show_orig=False)

结果:使用语义分割进行图像前景后景分离处理_第1张图片 

代码解析:

因为使用“cpu”确实太慢,所以实验环境改为google的colab,colab提供gpu支持,如果你没有gpu,则将代码中的“cuda”,改成“cpu”即可。

首先一开始,先将google云盘链接到实验环境中。

th, alpha = cv2.threshold(np.array(rgb), 0, 255, cv2.THRESH_BINARY),将rgb图像进行二值化处理得到alpha,alpha是一个由0和255组成的三通道图像,th是阈值,也就是这里的0.

alpha = cv2.GaussianBlur(alpha, (7, 7), 0)使用高斯平滑滤波处理,使图像不会太突兀。

# Multiply the foreground with the alpha matte

 foreground = cv2.multiply(alpha, foreground) 保留前景
# Multiply the background with ( 1 - alpha )
 background = cv2.multiply(1.0 - alpha, background) 保留背景
 # Add the masked foreground and background
outImage = cv2.add(foreground, background) 前景和背景混合

plt.imshow的输入图像如果是float型的列表,则元素取值范围必须是[0,1],如果是int型的列表,则元素取值范围必须是[0,255].。所以    return outImage.astype(int) 或 return outImage / 255 都是可以的。

2、改变背景

先上代码:

import os
from google.colab import drive
drive.mount('/content/drive')
path = "/content/drive/My Drive/Colab Notebooks"
os.chdir(path)
os.listdir(path)
from torchvision import models
from PIL import Image
import matplotlib.pyplot as plt
import torch
import numpy as np
import cv2

# Apply the transformations needed
import torchvision.transforms as T

def decode_segmap(image, source, bgimg, nc=21):
    label_colors = np.array([(0, 0, 0),  # 0=background
                             # 1=aeroplane, 2=bicycle, 3=bird, 4=boat, 5=bottle
                             (128, 0, 0), (0, 128, 0), (128, 128, 0), (0, 0, 128), (128, 0, 128),
                             # 6=bus, 7=car, 8=cat, 9=chair, 10=cow
                             (0, 128, 128), (128, 128, 128), (64, 0, 0), (192, 0, 0), (64, 128, 0),
                             # 11=dining table, 12=dog, 13=horse, 14=motorbike, 15=person
                             (192, 128, 0), (64, 0, 128), (192, 0, 128), (64, 128, 128), (192, 128, 128),
                             # 16=potted plant, 17=sheep, 18=sofa, 19=train, 20=tv/monitor
                             (0, 64, 0), (128, 64, 0), (0, 192, 0), (128, 192, 0), (0, 64, 128)])

    r = np.zeros_like(image).astype(np.uint8)
    g = np.zeros_like(image).astype(np.uint8)
    b = np.zeros_like(image).astype(np.uint8)

    for l in range(0, nc):
        idx = image == l
        r[idx] = label_colors[l, 0]
        g[idx] = label_colors[l, 1]
        b[idx] = label_colors[l, 2]

    rgb = np.stack([r, g, b], axis=2)

    # Load the foreground input image
    foreground = cv2.imread(source)

    # Load the background input image
    background = cv2.imread(bgimg)

    # Change the color of foreground image to RGB
    # and resize images to match shape of R-band in RGB output map
    foreground = cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB)
    background = cv2.cvtColor(background, cv2.COLOR_BGR2RGB)
    foreground = cv2.resize(foreground, (r.shape[1], r.shape[0]))
    background = cv2.resize(background, (r.shape[1], r.shape[0]))

    # Convert uint8 to float
    foreground = foreground.astype(float)
    background = background.astype(float)

    # Create a binary mask of the RGB output map using the threshold value 0
    th, alpha = cv2.threshold(np.array(rgb), 0, 255, cv2.THRESH_BINARY)

    # Apply a slight blur to the mask to soften edges
    alpha = cv2.GaussianBlur(alpha, (7, 7), 0)

    # Normalize the alpha mask to keep intensity between 0 and 1
    alpha = alpha.astype(float) / 255

    # Multiply the foreground with the alpha matte
    foreground = cv2.multiply(alpha, foreground)

    # Multiply the background with ( 1 - alpha )
    background = cv2.multiply(1.0 - alpha, background)

    # Add the masked foreground and background
    outImage = cv2.add(foreground, background)

    # Return a normalized output image for display
    return outImage / 255


def segment(net, path, bgimagepath, show_orig=True, dev='cuda'):
    img = Image.open(path)

    if show_orig: plt.imshow(img); plt.axis('off'); plt.show()
    # Comment the Resize and CenterCrop for better inference results
    trf = T.Compose([T.Resize(400),
                     # T.CenterCrop(224),
                     T.ToTensor(),
                     T.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])])
    inp = trf(img).unsqueeze(0).to(dev)
    out = net.to(dev)(inp)['out']
    om = torch.argmax(out.squeeze(), dim=0).detach().cpu().numpy()

    rgb = decode_segmap(om, path, bgimagepath)

    plt.imshow(rgb);
    plt.axis('off');
    plt.show()

dlab = models.segmentation.deeplabv3_resnet101(pretrained=1).eval()
segment(dlab, 'girl.jpg','bg.jpg', show_orig=False)

结果:

使用语义分割进行图像前景后景分离处理_第2张图片

原图

使用语义分割进行图像前景后景分离处理_第3张图片

背景图

使用语义分割进行图像前景后景分离处理_第4张图片

3、模糊背景

先上代码,注意:重复代码未包括。

def decode_segmap(image, source, nc=21):
  
  label_colors = np.array([(0, 0, 0),  # 0=background
               # 1=aeroplane, 2=bicycle, 3=bird, 4=boat, 5=bottle
               (128, 0, 0), (0, 128, 0), (128, 128, 0), (0, 0, 128), (128, 0, 128),
               # 6=bus, 7=car, 8=cat, 9=chair, 10=cow
               (0, 128, 128), (128, 128, 128), (64, 0, 0), (192, 0, 0), (64, 128, 0),
               # 11=dining table, 12=dog, 13=horse, 14=motorbike, 15=person
               (192, 128, 0), (64, 0, 128), (192, 0, 128), (64, 128, 128), (192, 128, 128),
               # 16=potted plant, 17=sheep, 18=sofa, 19=train, 20=tv/monitor
               (0, 64, 0), (128, 64, 0), (0, 192, 0), (128, 192, 0), (0, 64, 128)])

  r = np.zeros_like(image).astype(np.uint8)
  g = np.zeros_like(image).astype(np.uint8)
  b = np.zeros_like(image).astype(np.uint8)
  
  for l in range(0, nc):
    idx = image == l
    r[idx] = label_colors[l, 0]
    g[idx] = label_colors[l, 1]
    b[idx] = label_colors[l, 2]
    
    
  rgb = np.stack([r, g, b], axis=2)
  
  # Load the foreground input image 
  foreground = cv2.imread(source)

  # Change the color of foreground image to RGB 
  # and resize image to match shape of R-band in RGB output map  
  foreground = cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB)
  foreground = cv2.resize(foreground,(r.shape[1],r.shape[0]))

  # Create a Gaussian blur of kernel size 7 for the background image
  blurredImage = cv2.GaussianBlur(foreground, (7,7), 0)

  # Convert uint8 to float
  foreground = foreground.astype(float)
  blurredImage = blurredImage.astype(float)

  # Create a binary mask of the RGB output map using the threshold value 0
  th, alpha = cv2.threshold(np.array(rgb),0,255, cv2.THRESH_BINARY)

  # Apply a slight blur to the mask to soften edges
  alpha = cv2.GaussianBlur(alpha, (7,7),0)

  # Normalize the alpha mask to keep intensity between 0 and 1
  alpha = alpha.astype(float)/255

  # Multiply the foreground with the alpha matte
  foreground = cv2.multiply(alpha, foreground)  
  
  # Multiply the background with ( 1 - alpha )
  background = cv2.multiply(1.0 - alpha, blurredImage)  
  
  # Add the masked foreground and background
  outImage = cv2.add(foreground, background)

  # Return a normalized output image for display
  return outImage/255

def segment(net, path, show_orig=True, dev='cuda'):
  img = Image.open(path)
  
  if show_orig: plt.imshow(img); plt.axis('off'); plt.show()
  # Comment the Resize and CenterCrop for better inference results
  trf = T.Compose([T.Resize(450), 
                   #T.CenterCrop(224), 
                   T.ToTensor(), 
                   T.Normalize(mean = [0.485, 0.456, 0.406], 
                               std = [0.229, 0.224, 0.225])])
  inp = trf(img).unsqueeze(0).to(dev)
  out = net.to(dev)(inp)['out']
  om = torch.argmax(out.squeeze(), dim=0).detach().cpu().numpy()
  
  rgb = decode_segmap(om, path)
    
  plt.imshow(rgb); plt.axis('off'); plt.show()
segment(dlab, 'girl.png', show_orig=False)

结果:

使用语义分割进行图像前景后景分离处理_第5张图片

4、背景灰度化

上代码:

def decode_segmap(image, source, nc=21):
  
  label_colors = np.array([(0, 0, 0),  # 0=background
               # 1=aeroplane, 2=bicycle, 3=bird, 4=boat, 5=bottle
               (128, 0, 0), (0, 128, 0), (128, 128, 0), (0, 0, 128), (128, 0, 128),
               # 6=bus, 7=car, 8=cat, 9=chair, 10=cow
               (0, 128, 128), (128, 128, 128), (64, 0, 0), (192, 0, 0), (64, 128, 0),
               # 11=dining table, 12=dog, 13=horse, 14=motorbike, 15=person
               (192, 128, 0), (64, 0, 128), (192, 0, 128), (64, 128, 128), (192, 128, 128),
               # 16=potted plant, 17=sheep, 18=sofa, 19=train, 20=tv/monitor
               (0, 64, 0), (128, 64, 0), (0, 192, 0), (128, 192, 0), (0, 64, 128)])

  r = np.zeros_like(image).astype(np.uint8)
  g = np.zeros_like(image).astype(np.uint8)
  b = np.zeros_like(image).astype(np.uint8)
  
  for l in range(0, nc):
    idx = image == l
    r[idx] = label_colors[l, 0]
    g[idx] = label_colors[l, 1]
    b[idx] = label_colors[l, 2]
    
    
  rgb = np.stack([r, g, b], axis=2)

  # Load the foreground input image 
  foreground = cv2.imread(source)

  # Change the color of foreground image to RGB 
  # and resize image to match shape of R-band in RGB output map  
  foreground = cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB)
  foreground = cv2.resize(foreground,(r.shape[1],r.shape[0]))
  
  # Create a background image by copying foreground and converting into grayscale
  background = cv2.cvtColor(foreground, cv2.COLOR_BGR2GRAY)
  
  # convert single channel grayscale image to 3-channel grayscale image
  background = cv2.cvtColor(background, cv2.COLOR_GRAY2RGB)
  
  # Convert uint8 to float
  foreground = foreground.astype(float)
  background = background.astype(float)

  # Create a binary mask of the RGB output map using the threshold value 0
  th, alpha = cv2.threshold(np.array(rgb),0,255, cv2.THRESH_BINARY)

  # Apply a slight blur to the mask to soften edges
  alpha = cv2.GaussianBlur(alpha, (7,7),0)

  # Normalize the alpha mask to keep intensity between 0 and 1
  alpha = alpha.astype(float)/255

  # Multiply the foreground with the alpha matte
  foreground = cv2.multiply(alpha, foreground)  
  
  # Multiply the background with ( 1 - alpha )
  background = cv2.multiply(1.0 - alpha, background)  
  
  # Add the masked foreground and background
  outImage = cv2.add(foreground, background)

  # Return a normalized output image for display
  return outImage/255

def segment(net, path, show_orig=True, dev='cuda'):
  img = Image.open(path)
  
  if show_orig: plt.imshow(img); plt.axis('off'); plt.show()
  # Comment the Resize and CenterCrop for better inference results
  trf = T.Compose([T.Resize(450), 
                   #T.CenterCrop(224), 
                   T.ToTensor(), 
                   T.Normalize(mean = [0.485, 0.456, 0.406], 
                               std = [0.229, 0.224, 0.225])])
  inp = trf(img).unsqueeze(0).to(dev)
  out = net.to(dev)(inp)['out']
  om = torch.argmax(out.squeeze(), dim=0).detach().cpu().numpy()
  
  rgb = decode_segmap(om, path)
    
  plt.imshow(rgb); plt.axis('off'); plt.show()
  

dlab = models.segmentation.deeplabv3_resnet101(pretrained=1).eval()

segment(dlab, 'girl.jpg', show_orig=False)

使用语义分割进行图像前景后景分离处理_第6张图片

 

注意:本例中的不管是人还是车,标签对应的颜色,3个通道的值均不能为0,否则二值化掩码的时候就会丢失某些通道信息。

参考链接:https://www.learnopencv.com/applications-of-foreground-background-separation-with-semantic-segmentation/

你可能感兴趣的:(语义分割,PyTorch,原创)