如何在pyqt中通过OpenCV实现对窗口的透视变换

窗口的透视变换效果

当我们点击UWP应用中的小部件时，会发现小部件会朝着鼠标点击位置凹陷下去，而且不同的点击位置对应着不同的凹陷情况，看起来就好像小部件在屏幕上不只有x轴和y轴，甚至还有一个z轴。要做到这一点，其实只要对窗口进行透视变换即可。下面是对Qt的窗口和按钮进行透视变换的效果：

具体代码

PixmapPerspectiveTransform 类

它的作用是将传入的 QPixmap 转换为numpy 数组，然后用 opencv 的 warpPerspective 对数组进行透视变换，最后再将 numpy 数组转为 QPixmap 并返回；

# coding:utf-8

import cv2 as cv
import numpy
from PyQt5.QtGui import QImage, QPixmap

class PixmapPerspectiveTransform:
    """ 透视变换基类 """

    def __init__(self, pixmap=None):
        self.pixmap = pixmap

    def setPixmap(self, pixmap: QPixmap):
        """ 设置被变换的QPixmap """
        self.pixmap = QPixmap
        self.src=self.transQPixmapToNdarray(pixmap)
        self.height, self.width = self.src.shape[:2]
        # 变换前后的边角坐标
        self.srcPoints = numpy.float32(
            [[0, 0], [self.width - 1, 0], [0, self.height - 1],
             [self.width - 1, self.height - 1]])

    def setDstPoints(self, leftTop: list, rightTop, leftBottom, rightBottom):
        """ 设置变换后的边角坐标 """
        self.dstPoints = numpy.float32(
            [leftTop, rightTop, leftBottom, rightBottom])

    def getPerspectiveTransform(self, imWidth, imHeight, borderMode=cv.BORDER_CONSTANT, borderValue=[255, 255, 255, 0]) -> QPixmap:
        """ 透视变换图像，返回QPixmap

        Parameters
        ----------
        imWidth: int
            变换后的图像宽度

        imHeight: int
            变换后的图像高度

        borderMode: int
            边框插值方式

        borderValue: list
            边框颜色
        """
        # 如果是jpg需要加上一个透明通道
        if self.src.shape[-1] == 3:
            self.src = cv.cvtColor(self.src, cv.COLOR_BGR2BGRA)
        # 透视变换矩阵
        perspectiveMatrix = cv.getPerspectiveTransform(
            self.srcPoints, self.dstPoints)
        # 执行变换
        self.dst = cv.warpPerspective(self.src, perspectiveMatrix, (
            imWidth, imHeight), borderMode=borderMode, borderValue=borderValue)
        # 将ndarray转换为QPixmap
        return self.transNdarrayToQPixmap(self.dst)

    def transQPixmapToNdarray(self, pixmap: QPixmap):
        """ 将QPixmap转换为numpy数组 """
        width, height = pixmap.width(), pixmap.height()
        channels_count = 4
        image = pixmap.toImage()  # type:QImage
        s = image.bits().asstring(height * width * channels_count)
        # 得到BGRA格式数组
        array = numpy.fromstring(s, numpy.uint8).reshape(
            (height, width, channels_count))
        return array

    def transNdarrayToQPixmap(self, array):
        """ 将numpy数组转换为QPixmap """
        height, width, bytesPerComponent = array.shape
        bytesPerLine = 4 * width
        # 默认数组维度为 m*n*4
        dst = cv.cvtColor(array, cv.COLOR_BGRA2RGBA)
        pix = QPixmap.fromImage(
            QImage(dst.data, width, height, bytesPerLine, QImage.Format_RGBA8888))
        return pix

PerspectiveWidget 类

当我们的鼠标单击这个类实例化出来的窗口时，窗口会先通过 self.grab() 被渲染为QPixmap，然后调用 PixmapPerspectiveTransform 中的方法对QPixmap进行透视变换，拿到透视变换的结果后只需隐藏窗口内的小部件并通过 PaintEvent 将结果绘制到窗口上即可。虽然思路很通顺，但是实际操作起来会发现对于透明背景的窗口进行透视变换时，与透明部分交界的部分会被插值上半透明的像素。对于本来就属于深色的像素来说这没什么，但是如果像素是浅色的就会带来很大的视觉干扰，你会发现这些浅色部分旁边被描上了一圈黑边，我们先将这个图像记为img_1。img_1 差不多长这个样子，可以很明显看出白色的文字围绕着一圈黑色的描边。

为了解决这个烦人的问题，我又对桌面上的窗口进行截屏，再次透视变换。注意是桌面上看到的窗口，这时的窗口肯定是会有背景的，这时的透视变换就不会存在上述问题，记这个透视变换完的图像为 img_2。但实际上我们本来是不想要 img_2 中的背景的，所以只要将 img_2 中的背景替换完img_1中的透明背景，下面是具体代码：

# coding:utf-8

import numpy as np

from PyQt5.QtCore import QPoint, Qt
from PyQt5.QtGui import QPainter, QPixmap, QScreen, QImage
from PyQt5.QtWidgets import QApplication, QWidget

from my_functions.get_pressed_pos import getPressedPos
from my_functions.perspective_transform_cv import PixmapPerspectiveTransform

class PerspectiveWidget(QWidget):
    """ 可进行透视变换的窗口 """

    def __init__(self, parent=None, isTransScreenshot=False):
        super().__init__(parent)
        self.__visibleChildren = []
        self.__isTransScreenshot = isTransScreenshot
        self.__perspectiveTrans = PixmapPerspectiveTransform()
        self.__screenshotPix = None
        self.__pressedPix = None
        self.__pressedPos = None

    @property
    def pressedPos(self) -> str:
        """ 返回鼠标点击位置 """
        return self.__pressedPos

    def mousePressEvent(self, e):
        """ 鼠标点击窗口时进行透视变换 """
        super().mousePressEvent(e)
        # 多次点击时不响应，防止小部件的再次隐藏
        if self.__pressedPos:
            return
        self.grabMouse()
        pixmap = self.grab()
        self.__perspectiveTrans.setPixmap(pixmap)
        # 根据鼠标点击位置的不同设置背景封面的透视变换
        self.__setDstPointsByPressedPos(getPressedPos(self,e))
        # 获取透视变换后的QPixmap
        self.__pressedPix = self.__getTransformPixmap()
        # 对桌面上的窗口进行截图
        if self.__isTransScreenshot:
            self.__adjustTransformPix()
        # 隐藏本来看得见的小部件
        self.__visibleChildren = [
            child for child in self.children() if hasattr(child, 'isVisible') and child.isVisible()]
        for child in self.__visibleChildren:
            if hasattr(child, 'hide'):
                child.hide()
        self.update()

    def mouseReleaseEvent(self, e):
        """ 鼠标松开时显示小部件 """
        super().mouseReleaseEvent(e)
        self.releaseMouse()
        self.__pressedPos = None
        self.update()
        # 显示小部件
        for child in self.__visibleChildren:
            if hasattr(child, 'show'):
                child.show()

    def paintEvent(self, e):
        """ 绘制背景 """
        super().paintEvent(e)
        painter = QPainter(self)
        painter.setRenderHints(QPainter.Antialiasing | QPainter.HighQualityAntialiasing |
                               QPainter.SmoothPixmapTransform)
        painter.setPen(Qt.NoPen)
        # 绘制背景图片
        if self.__pressedPos:
            painter.drawPixmap(self.rect(), self.__pressedPix)

    def __setDstPointsByPressedPos(self,pressedPos:str):
        """ 通过鼠标点击位置设置透视变换的四个边角坐标 """
        self.__pressedPos = pressedPos
        if self.__pressedPos == 'left':
            self.__perspectiveTrans.setDstPoints(
                [5, 4], [self.__perspectiveTrans.width - 2, 1],
                [3, self.__perspectiveTrans.height - 3],
                [self.__perspectiveTrans.width - 2, self.__perspectiveTrans.height - 1])
        elif self.__pressedPos == 'left-top':
            self.__perspectiveTrans.setDstPoints(
                [7, 6], [self.__perspectiveTrans.width - 1, 1],
                [1, self.__perspectiveTrans.height - 2],
                [self.__perspectiveTrans.width - 2, self.__perspectiveTrans.height - 1])
        elif self.__pressedPos == 'left-bottom':
            self.__perspectiveTrans.setDstPoints(
                [0, 1], [self.__perspectiveTrans.width - 3, 0],
                [6, self.__perspectiveTrans.height - 5],
                [self.__perspectiveTrans.width - 2, self.__perspectiveTrans.height - 2])
        elif self.__pressedPos == 'top':
            self.__perspectiveTrans.setDstPoints(
                [4, 5], [self.__perspectiveTrans.width - 5, 5],
                [0, self.__perspectiveTrans.height - 1],
                [self.__perspectiveTrans.width - 1, self.__perspectiveTrans.height - 1])
        elif self.__pressedPos == 'center':
            self.__perspectiveTrans.setDstPoints(
                [3, 4], [self.__perspectiveTrans.width - 4, 4],
                [3, self.__perspectiveTrans.height - 3],
                [self.__perspectiveTrans.width - 4, self.__perspectiveTrans.height - 3])
        elif self.__pressedPos == 'bottom':
            self.__perspectiveTrans.setDstPoints(
                [0, 0], [self.__perspectiveTrans.width - 1, 0],
                [4, self.__perspectiveTrans.height - 4],
                [self.__perspectiveTrans.width - 5, self.__perspectiveTrans.height - 4])
        elif self.__pressedPos == 'right-bottom':
            self.__perspectiveTrans.setDstPoints(
                [1, 0], [self.__perspectiveTrans.width - 3, 2],
                [1, self.__perspectiveTrans.height - 2],
                [self.__perspectiveTrans.width - 6, self.__perspectiveTrans.height - 5])
        elif self.__pressedPos == 'right-top':
            self.__perspectiveTrans.setDstPoints(
                [0, 1], [self.__perspectiveTrans.width - 7, 5],
                [2, self.__perspectiveTrans.height - 1],
                [self.__perspectiveTrans.width - 2, self.__perspectiveTrans.height - 2])
        elif self.__pressedPos == 'right':
            self.__perspectiveTrans.setDstPoints(
                [1, 1], [self.__perspectiveTrans.width - 6, 4],
                [2, self.__perspectiveTrans.height - 1],
                [self.__perspectiveTrans.width - 4, self.__perspectiveTrans.height - 3])

    def __getTransformPixmap(self) -> QPixmap:
        """ 获取透视变换后的QPixmap """
        pix = self.__perspectiveTrans.getPerspectiveTransform(
            self.__perspectiveTrans.width, self.__perspectiveTrans.height).scaled(
                self.size(), Qt.KeepAspectRatio, Qt.SmoothTransformation)
        return pix

    def __getScreenShot(self) -> QPixmap:
        """ 对窗口口所在的桌面区域进行截图 """
        screen = QApplication.primaryScreen()  # type:QScreen
        pos = self.mapToGlobal(QPoint(0, 0))   # type:QPoint
        pix = screen.grabWindow(
            0, pos.x(), pos.y(), self.width(), self.height())
        return pix

    def __adjustTransformPix(self):
        """ 对窗口截图再次进行透视变换并将两张图融合，消除可能存在的黑边 """
        self.__screenshotPix = self.__getScreenShot()
        self.__perspectiveTrans.setPixmap(self.__screenshotPix)
        self.__screenshotPressedPix = self.__getTransformPixmap()
        # 融合两张透视图
        img_1 = self.__perspectiveTrans.transQPixmapToNdarray(self.__pressedPix)
        img_2 = self.__perspectiveTrans.transQPixmapToNdarray(self.__screenshotPressedPix)
        # 去除非透明背景部分
        mask = img_1[:, :, -1] == 0
        img_2[mask] = img_1[mask]
        self.__pressedPix = self.__perspectiveTrans.transNdarrayToQPixmap(img_2)

在mousePressEvent中调用了一个全局函数 getPressedPos(widget,e) ，如果将窗口分为九宫格，它就是用来获取判断鼠标的点击位置落在九宫格的哪个格子的，因为我在其他地方有用到它，所以没将其设置为PerspectiveWidget的方法成员。下面是这个函数的代码：

# coding:utf-8

from PyQt5.QtGui import QMouseEvent

def getPressedPos(widget, e: QMouseEvent) -> str:
    """ 检测鼠标并返回按下的方位 """
    pressedPos = None
    width = widget.width()
    height = widget.height()
    leftX = 0 <= e.x() <= int(width / 3)
    midX = int(width / 3) < e.x() <= int(width * 2 / 3)
    rightX = int(width * 2 / 3) < e.x() <= width
    topY = 0 <= e.y() <= int(height / 3)
    midY = int(height / 3) < e.y() <= int(height * 2 / 3)
    bottomY = int(height * 2 / 3) < e.y() <= height
    # 获取点击位置
    if leftX and topY:
        pressedPos = 'left-top'
    elif midX and topY:
        pressedPos = 'top'
    elif rightX and topY:
        pressedPos = 'right-top'
    elif leftX and midY:
        pressedPos = 'left'
    elif midX and midY:
        pressedPos = 'center'
    elif rightX and midY:
        pressedPos = 'right'
    elif leftX and bottomY:
        pressedPos = 'left-bottom'
    elif midX and bottomY:
        pressedPos = 'bottom'
    elif rightX and bottomY:
        pressedPos = 'right-bottom'
    return pressedPos

使用方法

很简单，只要将代码中的 QWidget 替换为 PerspectiveWidget。要对按钮也进行透视变换，只要按代码中所做的那样重写mousePressEvent、mouseReleaseEvent 和 paintEvent 即可，如果有对按钮使用qss，记得在 paintEvent 中加上super().paintEvent(e)，这样样式表才会起作用。总之框架已经给出，具体操作取决于你。如果你喜欢这篇博客的话，记得点个赞哦~~