【手动配准】python/opencv 手动选取特征点进行图像配准（图像可滚轮缩放、鼠标拖动）
背景
近日在做图像配准相关算法的研究，需要手动地配准图像，并将单应性矩阵保存为numpy数组，作为Ground truth验证配准算法特征点匹配地准确率。
该博主完成了我想实现大部分功能，但他使用的是C++，无法满足我需要保存numpy数组的需求，参考该博客自己写了一个python版本的手动配准代码。
主要功能
- 弹窗选择待配准图像
- 滚轮缩放图像
- 鼠标左键拖拽图像
- 鼠标右键点选特征点
- 匹配效果验证
代码
from pathlib import Path
import cv2
import tkinter as tk
from tkinter import filedialog
import tkinter.messagebox as GUI
import numpy as np
# 窗口大小
g_window_wh = [800, 600]
class struct_getPoint:
def __init__(self, image, window):
self.location_click = [0, 0]
self.location_release = [0, 0]
self.image_original = image.copy()
self.image_show = self.image_original[0: g_window_wh[1], 0:g_window_wh[0]]
self.location_win = [0, 0]
self.location_win_click = [0, 0]
self.image_zoom = self.image_original.copy()
self.zoom = 1
self.step = 0.1
self.window_name = window
self.point = []
# OpenCV鼠标事件
def getPoint(self):
def mouse_callback(event, x, y, flags, param):
def check_location(img_wh, win_wh, win_xy):
for i in range(2):
if win_xy[i] < 0:
win_xy[i] = 0
elif win_xy[i] + win_wh[i] > img_wh[i] and img_wh[i] > win_wh[i]:
win_xy[i] = img_wh[i] - win_wh[i]
elif win_xy[i] + win_wh[i] > img_wh[i] and img_wh[i] < win_wh[i]:
win_xy[i] = 0
# print(img_wh, win_wh, win_xy)
# 计算缩放倍数
# flag：鼠标滚轮上移或下移的标识, step：缩放系数，滚轮每步缩放0.1, zoom：缩放倍数
def count_zoom(flag, step, zoom, zoom_max):
if flag > 0: # 滚轮上移
zoom += step
if zoom > 1 + step * 20: # 最多只能放大到3倍
zoom = 1 + step * 20
else: # 滚轮下移
zoom -= step
if zoom < zoom_max: # 最多只能缩小到0.1倍
zoom = zoom_max
zoom = round(zoom, 2) # 取2位有效数字
return zoom
if event or flags:
w2, h2 = g_window_wh # 窗口的宽高
h1, w1 = param.image_zoom.shape[0:2] # 缩放图片的宽高
if event == cv2.EVENT_LBUTTONDOWN: # 左键点击
param.location_click = [x, y] # 左键点击时，鼠标相对于窗口的坐标
param.location_win_click = [param.location_win[0],
param.location_win[1]] # 窗口相对于图片的坐标，不能写成location_win = g_location_win
elif event == cv2.EVENT_MOUSEMOVE and (flags & cv2.EVENT_FLAG_LBUTTON): # 按住左键拖曳
param.location_release = [x, y] # 左键拖曳时，鼠标相对于窗口的坐标
if w1 <= w2 and h1 <= h2: # 图片的宽高小于窗口宽高，无法移动
param.location_win = [0, 0]
elif w1 >= w2 and h1 < h2: # 图片的宽度大于窗口的宽度，可左右移动
param.location_win[0] = param.location_win_click[0] + param.location_click[0] - \
param.location_release[0]
elif w1 < w2 and h1 >= h2: # 图片的高度大于窗口的高度，可上下移动
param.location_win[1] = param.location_win_click[1] + param.location_click[1] - \
param.location_release[1]
else: # 图片的宽高大于窗口宽高，可左右上下移动
param.location_win[0] = param.location_win_click[0] + param.location_click[0] - \
param.location_release[0]
param.location_win[1] = param.location_win_click[1] + param.location_click[1] - \
param.location_release[1]
check_location([w1, h1], [w2, h2], param.location_win) # 矫正窗口在图片中的位置
elif event == cv2.EVENT_MOUSEWHEEL: # 滚轮
z = param.zoom # 缩放前的缩放倍数，用于计算缩放后窗口在图片中的位置
zoom_max = g_window_wh[0] / param.image_original.shape[1]
param.zoom = count_zoom(flags, param.step, param.zoom, zoom_max) # 计算缩放倍数
w1, h1 = [int(param.image_original.shape[1] * param.zoom),
int(param.image_original.shape[0] * param.zoom)] # 缩放图片的宽高
param.image_zoom = cv2.resize(param.image_original, (w1, h1), interpolation=cv2.INTER_AREA) # 图片缩放
param.location_win = [int((param.location_win[0] + x) * param.zoom / z - x),
int((param.location_win[1] + y) * param.zoom / z - y)] # 缩放后，窗口在图片的位置
check_location([w1, h1], [w2, h2], param.location_win) # 矫正窗口在图片中的位置
elif event == cv2.EVENT_RBUTTONDOWN: # 右键选点
point_num = len(param.point)
[x_ori, y_ori] = [int((param.location_win[0] + x) / param.zoom),
int((param.location_win[1] + y) / param.zoom)]
param.point.append([x_ori, y_ori])
cv2.circle(param.image_original, (x_ori, y_ori), 3, (255, 0, 0), thickness=-1) # 画圆半径为3，并填充
cv2.putText(param.image_original, str(point_num + 1), (x_ori, y_ori), cv2.FONT_HERSHEY_PLAIN,
1.0, (0, 255, 0), thickness=1) # 加入文字，位置，字体，尺度因子，颜色，粗细
param.image_zoom = cv2.resize(param.image_original, (w1, h1), interpolation=cv2.INTER_AREA) # 图片缩放
param.image_show = param.image_zoom[param.location_win[1]:param.location_win[1] + h2,
param.location_win[0]:param.location_win[0] + w2] # 实际的显示图片
cv2.imshow(param.window_name, param.image_show)
cv2.namedWindow(self.window_name, cv2.WINDOW_NORMAL)
cv2.resizeWindow(self.window_name, g_window_wh[0], g_window_wh[1])
cv2.imshow(self.window_name, self.image_show)
cv2.setMouseCallback(self.window_name, mouse_callback, self)
def getHomography():
while True:
root_path1 = filedialog.askopenfilename(title='请选择图片')
root_path2 = filedialog.askopenfilename(title='请选择图片')
img1 = cv2.imread(root_path1)
img2 = cv2.imread(root_path2)
imgIn1 = img1.copy()
imgIn2 = img2.copy()
# 点选特征点
C1 = struct_getPoint(img1, "window1")
C1.getPoint()
C2 = struct_getPoint(img2, "window2")
C2.getPoint()
cv2.waitKey(0) # 等待键盘点击事件来结束阻塞
point1 = C1.point
point2 = C2.point
# 判断特征点数量是否一致
if len(point1) != len(point2):
GUI.showinfo(title='提示', message='特征数量不一致，请重新选择')
continue
# 估计由输出坐标到输入坐标的单应性变换模型
Homo, error= cv2.findHomography(np.array(point1), np.array(point2))
# 产生输出图像
[H, W] = img2.shape[0:2]
imgOut = cv2.warpPerspective(img1, Homo, (W, H))
imgRes = cv2.addWeighted(img2, 0.5, imgOut, 0.5, 0)
cv2.namedWindow("result", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
cv2.imshow('result', imgRes)
cv2.waitKey(0)
cv2.destroyAllWindows() # 销毁所有窗口
if __name__ == '__main__':
root = tk.Tk()
root.withdraw()
getHomography()欢迎大家交流讨论！
本文章为转载内容，我们尊重原作者对文章享有的著作权。如有内容错误或侵权问题，欢迎原作者联系我们进行内容更正或删除文章。
