前言
当图中想传达的区域太小,而在不裁剪的情况下又想要放大图中的某一部分区域,实现放大镜的效果。


代码
为了能方便友友们直接上手使用,我编写了可交互式的局部放大功能的代码。大家只需要输入自己图像的路径就可以体验啦。
import cv2
import os
import numpy as np
img_path= r'' # 请输入自己需要放大图像的路径
img_name = os.path.basename(img_path)
img = cv2.imread(img_path)
def on_EVENT_LBUTTONDOWN(event, x, y, flags, param):
global img
if event == cv2.EVENT_LBUTTONDOWN: # 按下鼠标左键则放大所点的区域
xy = "%d,%d" % (x, y)
print (xy)
length = 20 # 局部区域的边长的一半
big_length = 200 # 放大后图像的边长
part_left = x - length
part_right = x + length
part_top = y - length
part_bottom = y + length
height, width, _ = np.shape(img)
if (x < width / 2) & (y < height / 2):
loc_left = 10
loc_top = 10
loc_right = loc_left + big_length
loc_bottom = loc_top + big_length
cv2.line(img,(part_right,part_top),(loc_right,loc_top),(0,0,0),2)
cv2.line(img,(part_left,part_bottom),(loc_left,loc_bottom),(0,0,0),2)
elif (x >= width / 2) & (y < height / 2):
loc_right = width - 10
loc_left = loc_right - big_length
loc_top = 10
loc_bottom = loc_top + big_length
cv2.line(img,(part_left,part_top),(loc_left,loc_top),(0,0,0),2)
cv2.line(img,(part_right,part_bottom),(loc_right,loc_bottom),(0,0,0),2)
elif (x < width / 2) & (y >= height / 2):
loc_left = 10
loc_right = loc_left + big_length
loc_bottom = height - 10
loc_top = loc_bottom - big_length
cv2.line(img,(part_left,part_top),(loc_left,loc_top),(0,0,0),2)
cv2.line(img,(part_right,part_bottom),(loc_right,loc_bottom),(0,0,0),2)
elif (x >= width / 2) & (y >= height / 2):
loc_bottom = height - 10
loc_top = loc_bottom - big_length
loc_right = width - 10
loc_left = loc_right - big_length
cv2.line(img,(part_right,part_top),(loc_right,loc_top),(0,0,0),2)
cv2.line(img,(part_left,part_bottom),(loc_left,loc_bottom),(0,0,0),2)
part = img[part_top:part_bottom,part_left:part_right]
mask = cv2.resize(part, (big_length, big_length), fx=0, fy=0, interpolation=cv2.INTER_LINEAR)
img[loc_top:loc_bottom,loc_left:loc_right]=mask
cv2.rectangle(img,(part_left,part_top),(part_right,part_bottom),(0,0,0),2)
cv2.rectangle(img,(loc_left,loc_top),(loc_right,loc_bottom),(0,0,0),2)
cv2.imshow("image", img)
if event == cv2.EVENT_RBUTTONDOWN: # 按下鼠标右键恢复原图
img = cv2.imread(img_path)
cv2.imshow("image", img)
cv2.namedWindow("image")
cv2.setMouseCallback("image", on_EVENT_LBUTTONDOWN)
cv2.imshow("image", img)
cv2.waitKey(0)
cv2.imwrite("image1.jpg", img)
以下是简介的版本,大家只需要修改某些参数就能实现上述功能啦。
import cv2
import os
import numpy as np
img_path= r'' # 请输入自己需要放大图像的路径
img = cv2.imread(img_path)
# 输入局部点的中心点的坐标以及截取的边长的一半
x = 364
y = 261
length = 20
part_left = x - length
part_right = x + length
part_top = y - length
part_bottom = y + length
# 输入放大后图像左上点位于原图中的位置和边长
loc_left = 10
loc_top = 10
big_length = 200
loc_right = loc_left + big_length
loc_bottom = loc_top + big_length
# 截取图像中的局部区域并放大
part = img[part_top:part_bottom,part_left:part_right]
mask = cv2.resize(part, (big_length, big_length), fx=0, fy=0, interpolation=cv2.INTER_LINEAR)
# 将放大后的图像放置位于指定区域
img[loc_top:loc_bottom,loc_left:loc_right]=mask
# 画一些图形引导人的视线(可选择)
cv2.rectangle(img,(part_left,part_top),(part_right,part_bottom),(0,0,0),2)
cv2.rectangle(img,(loc_left,loc_top),(loc_right,loc_bottom),(0,0,0),2)
cv2.line(img,(part_right,part_top),(loc_right,loc_top),(0,0,0),2)
cv2.line(img,(part_left,part_bottom),(loc_left,loc_bottom),(0,0,0),2)
# 展示结果并保存
cv2.namedWindow("image2")
cv2.imshow("image2", img)
cv2.imwrite("image3.jpg", img)
cv2.waitKey(0)
下面具体介绍一下这个代码的核心部分,方便大家后续进行魔改。
1.获取局部区域并插值放大
part = img[part_top:part_bottom,part_left:part_right]
mask = cv2.resize(part, (big_length, big_length), fx=0, fy=0, interpolation=cv2.INTER_LINEAR)
2.将放大的图像放在原图中的某个位置
img[loc_top:loc_bottom,loc_left:loc_right]=mask
3.可以添加边框和线指引人的视线
cv2.rectangle(img,(part_left,part_top),(part_right,part_bottom),(0,0,0),2)
cv2.rectangle(img,(loc_left,loc_top),(loc_right,loc_bottom),(0,0,0),2)
cv2.line(img,(part_right,part_top),(loc_right,loc_top),(0,0,0),2)
cv2.line(img,(part_left,part_bottom),(loc_left,loc_bottom),(0,0,0),2)
应用
如果大家也是一名研究生,可以利用这些代码制作以下类型的图片。
Green Fluorescent Protein and Phase Contrast Image Fusion Via Detail Preserving Cross Network
Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network