pyqt5笔记——截取QLabel图片四个顶点，做文档较正
要求
- 利用QLabel控件显示图片；
- 鼠标点击，获取单机坐标，根据四个顶点坐标，在图中画出截取区域；
- 确定，根据鼠标点击的四个点，求取单适应矩阵，做图像较正。
说明
画图
在QLabel画图，需要继承QLabel类，重写paintEvent和mouseReleaseEvent。
需要注意的是，在mouseReleaseEvent中获取鼠标点击的点的时候，global_point = event.globalPos()是在你1920*1080的屏幕上的点击坐标，event.pos()是你在UI窗口中的坐标，UI界面左上角为(0,0)，而用event.globalPos()获取UI界面左上角的坐标可能是(765,532)，一般不为0。而真实对应的图片中的坐标则是pos()-(centralWidget.pos()+leftImgLabel.pos())具体见参考3。
有一个简单的方法，获得真实图片中的坐标，利用self.mapFromGlobal(global_point)即可。
在ImageLabel类中，pantEvent和mouseReleaseEvent代码如下：
def paintEvent(self, event):
QLabel.paintEvent(self, event)
painter = QPainter()
painter.begin(self)
pen = QPen(Qt.red, 4 , Qt.DashDotLine)#虚线画笔
painter.setPen(pen)
for k in range(len(self.points)):
if k + 1 != len(self.points):
painter.drawLine(self.points[k][0], self.points[k][1], self.points[k + 1][0], self.points[k + 1][1])
else:
painter.drawLine(self.points[k][0], self.points[k][1], self.points[0][0], self.points[0][1])
painter.end()
# 开启标记功能时，获取点击坐标
def mouseReleaseEvent(self, event):
if len(self.points) < 4:
global_point = event.globalPos()
local_point = self.mapFromGlobal(global_point)
point_x = local_point.x()
point_y = local_point.y()
self.points.append([point_x, point_y])
self.update()#获取鼠标点击的点之后，通知画线功能如下：
求解单适应矩阵
可以用 h, s = cv2.findHomography(src_point, dst_point, cv2.RANSAC, 10) 来求解。
我们获取上面的点范围是(400， 600)之后，需要利用mapfromLoal()函数，将点击的坐标，还原到真实图像(3024, 4536)上，需要注意的是，要让上面两个坐标值，是按比例缩放，否则，会有问题。mapfromLoal()代码如下:
def mapfromLoal(self, points):
points_origanal = []
y_ratio = np.float32(self.image.shape[0] / self.image_label.height())
x_ratio = np.float32(self.image.shape[1] / self.image_label.width())
for point in points:
points_origanal.append([point[0] * x_ratio, point[1] * y_ratio])
return points_origanal单适应矩阵求解如下：
points = self.image_label.get_points()#获取image_label上点击的点
#单适应矩阵
points_origanal = self.mapfromLoal(points)
src_point = np.float32(points_origanal)
print("src_point", src_point)
#想要变换成图像的大小
dsize = (self.image.shape[1], self.image.shape[0])
#变换前后四个点要一一对应
dst_point = np.float32([[0, 0], [0, dsize[1] - 1], [dsize[0] - 1, dsize[1] - 1], [dsize[0] - 1, 0]])
print("dst_point", dst_point)
h, s = cv2.findHomography(src_point, dst_point, cv2.RANSAC, 10)
self.image = cv2.warpPerspective(self.image, h, dsize,borderMode=cv2.BORDER_REPLICATE)
cv2.imwrite("warp.jpg", self.image)以上只是简单的求解，点击4个顶点的顺序只能按照左上、左下，右下，右上的顺序来。下面修改一下mapfromLocal函数，对四个点进行排序，之后，不管如何点击，都可以准确的进行变换。参考4.的思路，写成代码，修改后的map函数如下：
def mapfromLoal(self, points):
#从局部点投影到原图，并且将4个点的顺序，按照左上、左下，右下，右上的顺序排序
points_origanal = []
print("in map shape:", self.image.shape)
print("before largen", points)
y_ratio = np.float32(self.image.shape[0] / self.image_label.height())
x_ratio = np.float32(self.image.shape[1] / self.image_label.width())
for point in points:
points_origanal.append([point[0] * x_ratio, point[1] * y_ratio])
order_points = self.order_points(points_origanal)
print("order",order_points)
return order_pointsorder_points()思路如下：
- 先计算出这四个点的中心位置，然后根据中心位置来做判断。
- 那么中心位置就是把他们四个点的坐标全部加起来再除以4。
- 与中心位置的x坐标比较，区分左右
- 对于左边的点，根据与中心位置的y坐标区分上下
- 对于右边的点，同理区分上下
- 按照左上，左下，右下，右上的顺序返回
代码如下：
#根据左上、左下，右下，右上的顺序排序
def order_points(self, points):
#求中心点的坐标
center = [0,0]
for point in points:
center[0] += point[0]
center[1] += point[1]
center[0] = center[0] / 4
center[1] = center[1] / 4
print(center)
#根据中心点x坐标，大于为左，小于为右
left = []
right = []
for point in points:
if point[0] > center[0]:
right.append(point)
else:
left.append(point)
print("left",left, "right",right)
#区分左边坐标的上下，y坐标大于中心坐标为下，否则为上
bl = []
tl = []
for l in left:
if l[1] > center[1]:
bl = l
else:
tl = l
#同理区分右上，右下
br = []
tr = []
for r in right:
if r[1] > center[1]:
br = r
else:
tr = r
return [tl, bl, br, tr]
全部代码
from PyQt5.QtWidgets import *
from PyQt5.QtCore import *
from PyQt5.QtGui import *
import sys
import os
import numpy as np
import threading
import cv2
import datetime
buffer = ''
class ImageLabel(QLabel):
'''获取用户裁剪坐标点，画线
Attributes:
points:用户点击的裁剪点
'''
def __init__(self, parent=None):
super(ImageLabel, self).__init__(parent)
self.points = []
#normal function
def show_image(self, image):
#参数image为np.array类型
rgb_image = cv2.cvtColor(image.copy(), cv2.COLOR_BGR2RGB)
rgb_image = cv2.resize(rgb_image, (self.width(), self.height()))
label_image = QImage(rgb_image.data, rgb_image.shape[1], rgb_image.shape[0], QImage.Format_RGB888)
self.setPixmap(QPixmap(label_image))
def reselect(self):
self.points.clear()
self.update()
def get_points(self):
return self.points
#slot function
# 根据点过的点来画图
def paintEvent(self, event):
QLabel.paintEvent(self, event)
painter = QPainter()
painter.begin(self)
pen = QPen(Qt.red, 4 , Qt.DashDotLine)#虚线画笔
painter.setPen(pen)
for k in range(len(self.points)):
if k + 1 != len(self.points):
painter.drawLine(self.points[k][0], self.points[k][1], self.points[k + 1][0], self.points[k + 1][1])
else:
painter.drawLine(self.points[k][0], self.points[k][1], self.points[0][0], self.points[0][1])
painter.end()
# 开启标记功能时，获取点击坐标
def mouseReleaseEvent(self, event):
if len(self.points) < 4:
global_point = event.globalPos()
local_point = self.mapFromGlobal(global_point)
point_x = local_point.x()
point_y = local_point.y()
self.points.append([point_x, point_y])
self.update()#获取鼠标点击的点之后，通知画线
class cropWindow(QWidget):
ok_signal = pyqtSignal()
def __init__(self, image):
super(cropWindow, self).__init__()
self.pannel_height = 400
self.pannel_width = 600
self.button_height = 64
self.button_width = 64
self.image = image
# 设置宽高比
w_h_ratio = self.pannel_width * 1.0 / self.pannel_height
print("w_h_ration",w_h_ratio)
self.image = cv2.resize(self.image, ((int)(self.image.shape[0] * w_h_ratio), self.image.shape[0]))
# self.image = image
self.image_label = ImageLabel(self)
self.image_label.setFixedHeight(self.pannel_height)
self.image_label.setFixedWidth(self.pannel_width)
self.image_label.show_image(self.image)
#button
self.ok_button = QPushButton(self)
self.ok_button.setFixedWidth(self.button_width)
self.ok_button.setFixedHeight(self.button_height)
self.ok_button.setToolTip("ok")
# self.ok_button.setText("ok")
self.ok_button.setIcon(QIcon("./icons/ok.png"))
self.ok_button.clicked.connect(self.on_ok_button)
self.reselect_button = QPushButton(self)
self.reselect_button.setFixedHeight(self.button_height)
self.reselect_button.setFixedWidth(self.button_width)
# self.reselect_button.setText("reset")
self.reselect_button.setIcon(QIcon("./icons/reset.png"))
self.reselect_button.setToolTip("cancel")
self.reselect_button.clicked.connect(self.on_reselect_button)
#layout
self.h_layout = QHBoxLayout()
self.h_layout.addWidget(self.ok_button)
self.h_layout.addWidget(self.reselect_button)
self.v_layout = QVBoxLayout()
self.v_layout.addWidget(self.image_label)
self.v_layout.addLayout(self.h_layout)
self.setLayout(self.v_layout)
self.setFixedWidth(630)
self.setFixedHeight(590)
#normal function
def get_image(self):
return self.image
#根据左上、左下，右下，右上的顺序排序
def order_points(self, points):
#求中心点的坐标
center = [0,0]
for point in points:
center[0] += point[0]
center[1] += point[1]
center[0] = center[0] / 4
center[1] = center[1] / 4
print(center)
#根据中心点x坐标，大于为左，小于为右
left = []
right = []
for point in points:
if point[0] > center[0]:
right.append(point)
else:
left.append(point)
print("left",left, "right",right)
#区分左边坐标的上下，y坐标大于中心坐标为下，否则为上
bl = []
tl = []
for l in left:
if l[1] > center[1]:
bl = l
else:
tl = l
#同理区分右上，右下
br = []
tr = []
for r in right:
if r[1] > center[1]:
br = r
else:
tr = r
return [tl, bl, br, tr]
def mapfromLoal(self, points):
#从局部点投影到原图，并且将4个点的顺序，按照左上、左下，右下，右上的顺序排序
points_origanal = []
print("in map shape:", self.image.shape)
print("before largen", points)
y_ratio = np.float32(self.image.shape[0] / self.image_label.height())
x_ratio = np.float32(self.image.shape[1] / self.image_label.width())
for point in points:
points_origanal.append([point[0] * x_ratio, point[1] * y_ratio])
order_points = self.order_points(points_origanal)
print("order",order_points)
return order_points
# return points_origanal
#slot function
def on_ok_button(self):
if len(self.image_label.get_points()) != 4:#判断是否选取完
reply = QMessageBox.warning(self,
"提示",
"请选取四个点",
QMessageBox.Ok)
return
reply = QMessageBox.warning(self,
"提示",
"你确定截取好了吗",
QMessageBox.Yes | QMessageBox.No)
if reply == QMessageBox.Yes:
points = self.image_label.get_points()
#单适应变换
points_origanal = self.mapfromLoal(points)
src_point = np.float32(points_origanal)
print("src_point", src_point)
#想要变换成图像的大小
dsize = (self.image.shape[1], self.image.shape[0])
dst_point = np.float32([[0, 0], [0, dsize[1] - 1], [dsize[0] - 1, dsize[1] - 1], [dsize[0] - 1, 0]])
print("dst_point", dst_point)
h, s = cv2.findHomography(src_point, dst_point, cv2.RANSAC, 10)
self.image = cv2.warpPerspective(self.image, h, dsize,borderMode=cv2.BORDER_REPLICATE)
cv2.imwrite("warp.jpg", self.image)
self.ok_signal.emit()
self.close()
def on_reselect_button(self):
self.image_label.reselect()
if __name__ == "__main__":
app = QApplication(sys.argv)
app.setApplicationName("较正")
window = cropWindow(cv2.imread("tests/2.jpg"))
window.show()
sys.exit(app.exec_())
参考
1.单应性(homography)变换的推导
2.计算单应矩阵，图像矫正，opencv的findHomography
3.Qt获取鼠标位置（绝对位置、相对位置）
4.根据四个点坐标排列出左上右上右下左下位置关系