如何让plt.imshow显示负坐标区域的变换后图像

问题说明 #

我正在用线性代数方法拼接两张图像成全景图，其中一张图像经变换后大部分内容处于小于0的坐标区域，导致plt.imshow无法显示这部分内容。想知道能不能扩展绘图范围展示负坐标区域，或者整体右移图像来呈现更多内容？试过expand和origin参数，但没成功。代码如下：

from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
img1 = Image.open("gwint1.png")
coeffs=np.array([0.207879489,0.1122809023,496.629559,-0.341172122,0.908759060,53.1000028,-0.000964805704,0.000223378490]).reshape(4,2)
plt.imshow(img1.transform((width,height),Image.PERSPECTIVE, coeffs, fill=0, fillcolor='black'),alpha=1);

三种可行解决方案 #

方案一：调整Matplotlib坐标轴范围，直接显示负坐标区域 #

plt.imshow默认只展示(0,0)到输出图像尺寸的坐标范围，你可以手动设置坐标轴的显示边界，把负坐标区域包含进来：

from PIL import Image
import matplotlib.pyplot as plt
import numpy as np

img1 = Image.open("gwint1.png")
width, height = img1.size  # 补全原代码中缺失的尺寸定义
coeffs=np.array([0.207879489,0.1122809023,496.629559,-0.341172122,0.908759060,53.1000028,-0.000964805704,0.000223378490]).reshape(4,2)

# 生成变换后的图像
transformed_img = img1.transform((width, height), Image.PERSPECTIVE, coeffs, fill=0, fillcolor='black')

plt.imshow(transformed_img, alpha=1)
# 根据你图像的实际负坐标范围调整数值，比如这里假设x最小到-500，y最小到-200
plt.xlim(-500, width)
plt.ylim(-200, height)
plt.show()

方案二：修改变换矩阵，整体偏移图像到正坐标区域（推荐用于全景拼接） #

先计算图像变换后的边界，然后修改透视变换矩阵，把整个图像平移到正坐标区域，这样后续拼接更方便：

from PIL import Image
import matplotlib.pyplot as plt
import numpy as np

img1 = Image.open("gwint1.png")
width, height = img1.size
coeffs = np.array([0.207879489, 0.1122809023, 496.629559,
                   -0.341172122, 0.908759060, 53.1000028,
                   -0.000964805704, 0.000223378490]).reshape(4, 2)

# 定义函数计算单个点的透视变换结果
def transform_point(point, coeffs):
    x, y = point
    a, b, c = coeffs[0], coeffs[1], coeffs[2]
    d, e, f = coeffs[3], coeffs[4], coeffs[5]
    g, h = coeffs[6], coeffs[7]
    denom = g * x + h * y + 1
    return (a*x + b*y + c)/denom, (d*x + e*y + f)/denom

# 计算原图像四个角变换后的坐标
corners = [(0,0), (width,0), (width,height), (0,height)]
transformed_corners = [transform_point(p, coeffs) for p in corners]

# 找到变换后最小的x和y值，计算需要偏移的量
min_x = min(p[0] for p in transformed_corners)
min_y = min(p[1] for p in transformed_corners)
offset_x = -min_x
offset_y = -min_y

# 修改变换矩阵的常数项，实现平移
new_coeffs = coeffs.copy()
new_coeffs[2] += offset_x  # 调整x方向的平移参数
new_coeffs[5] += offset_y  # 调整y方向的平移参数

# 计算新的输出尺寸，确保容纳整个图像
max_x = max(p[0] for p in transformed_corners) + offset_x
max_y = max(p[1] for p in transformed_corners) + offset_y
new_width = int(np.ceil(max_x))
new_height = int(np.ceil(max_y))

# 应用新变换
transformed_img = img1.transform((new_width, new_height), Image.PERSPECTIVE, new_coeffs, fill=0, fillcolor='black')

plt.imshow(transformed_img, alpha=1)
plt.show()

方案三：正确使用PIL的expand参数 #

你之前可能没正确用对expand参数，设置expand=True后，PIL会自动扩展输出图像尺寸，容纳变换后的全部内容：

from PIL import Image
import matplotlib.pyplot as plt
import numpy as np

img1 = Image.open("gwint1.png")
width, height = img1.size
coeffs=np.array([0.207879489,0.1122809023,496.629559,-0.341172122,0.908759060,53.1000028,-0.000964805704,0.000223378490]).reshape(4,2)

# 添加expand=True参数，自动扩展图像尺寸
transformed_img = img1.transform((width, height), Image.PERSPECTIVE, coeffs, fill=0, fillcolor='black', expand=True)

plt.imshow(transformed_img, alpha=1)
plt.show()

内容的提问来源于stack exchange，提问作者Tymoteusz Dziejma