写在前面：本博客仅作记录学习之用，部分图片来自网络，如需使用请注明出处，同时如有侵犯您的权益，请联系删除！
文章目录- 前言
- 局部放大
- 函数实现
- 单一局部放大置于图像内部
- 多局部放大拼接于图形侧面
- 主函数
- 效果
- 总结
- 致谢
本博客仅为学习记录之用，目的在于后续若需要相关的有资可查。在言语上恐有诸多纰漏，如有错误，欢迎指出交流学习！
本博客所包含的大致内容: 图片某一局部放大并置于图形内部；图片多个局部放大并拼接在原图侧面。
局部放大
为什么要局部放大？无论的图像生成、修复、去污等等，如生成对抗方法；亦或是图像分割，如unet。对于方法的结果都需要进行有效的展示，凸显方法的优越性能，往往细节的部分才是我们所更加关注的。但往往在图像本身很难观察到这些细节，因此有意识的将需要的局部放大是展示效果的有效方法。
函数实现
单一局部放大置于图像内部
思路也简单，首先获取到需要关注的部分，然后在将其进行插值放大，最后放在原图的某一位置即可。
本函数参数传入图像，需要目标区域的指标，其次放在的位置，支持四个角以及中心，以及放大的倍数。
目标区域为矩形，长宽比任意。
def Partial_magnification(pic, target, location='lower_right', ratio=1):
'''
:param pic: input pic
:param target: Intercept area, for example [target_x, target_y, target_w, target_h]
:param location: lower_right,lower_left,top_right,top_left,center
:param ratio: gain
:return: oringal pic, pic
'''
img = copy.copy(pic)
w, h = pic.shape[1], pic.shape[0],
target_x, target_y = target[0], target[1]
target_w, target_h = target[2], target[3]
cv2.rectangle(pic, (target_x, target_y), (target_x + target_w, target_y + target_h), (255, 255, 0), 2)
new_pic = pic[target_y:target_y + target_h, target_x:target_x + target_w]
new_pic = cv2.resize(new_pic, (target_w*ratio, target_h*ratio), interpolation=cv2.INTER_CUBIC)
if location == 'lower_right':
pic[h-1-target_h*ratio:h-1, w-1-target_w*ratio:w-1] = new_pic
cv2.line(pic, (target_x + target_w, target_y + target_h), (w-1-target_w*ratio, h-1-target_h*ratio), (255, 0, 0),2)
elif location == 'lower_left':
pic[h-1-target_h*ratio:h-1, 0:target_w*ratio] = new_pic
elif location == 'top_right':
pic[0:target_h*ratio, w-1-target_w*ratio:w-1] = new_pic
elif location == 'top_left':
pic[0:target_h*ratio, 0:target_w*ratio] = new_pic
elif location == 'center':
pic[int(h/2-target_h*ratio/2):int(h/2+target_h*ratio/2),
int(w/2-target_w*ratio/2):int(w/2+target_w*ratio/2)] = new_pic
return img, pic
多局部放大拼接于图形侧面
当存在多个局部需要放大时，若都置于原图内部，遮挡图片内容的可能性较大，因此将其置于图像侧面。
本函数仅支持两个局部区域放大。和第一个函数不同的是两个局部放大的区域长宽比需要一致，为了方便拼接。没有设置放大倍数，默认局部放大后拼接图片与原图等高。
def Partial_magnification_2(pic, target1, target2, *args):
'''
:param pic: input pic
:param target: Intercept area, for example [target_x, target_y, target_w, target_w]
:param location: lower_right,lower_left,top_right,top_left,center
:param ratio: gain
:return: oringal pic, pic
'''
# imgg = copy.copy(pic)
w, h = pic.shape[1], pic.shape[0],
assert target1[0]+target1[2] < w and target1[1]+target1[3] < h,
'The target1 area is too large and exceeds the image size'
assert target2[0]+target2[2] < w and target2[1]+target2[3] < h,
'The target2 area is too large and exceeds the image size'
assert target1[2] > 10 or target1[3] > 10,
'The target1 area is too small, not recommended'
assert target2[2] > 10 or target2[3] > 10,
'The target2 area is too small, not recommended'
assert target1[0] > 0 and target1[0] < w,
'The starting point of the target1 area is beyond the scope of the image'
assert target2[0] > 0 and target2[0] < w,
'The starting point of the target2 area is beyond the scope of the image'
if target2[2] / target2[3] == target1[2] / target1[3]: #
R = target2[2]/target2[3]
if target1[1] > target2[1]:
target1_x, target1_y = target2[0], target2[1]
target1_w, target1_h = target2[2], target2[3]
target2_x, target2_y = target1[0], target1[1]
target2_w, target2_h = target1[2], target1[3]
else:
target1_x, target1_y = target1[0], target1[1]
target1_w, target1_h = target1[2], target1[3]
target2_x, target2_y = target2[0], target2[1]
target2_w, target2_h = target2[2], target2[3]
cv2.rectangle(pic, (target1_x, target1_y), (target1_x + target1_w, target1_y + target1_h), (255, 250, 255), 2)
cv2.rectangle(pic, (target2_x, target2_y), (target2_x + target2_w, target2_y + target2_h), (255, 252, 255), 2)
new_pic1 = pic[target1_y:target1_y + target1_h, target1_x:target1_x + target1_w]
new_pic1 = cv2.resize(new_pic1, (int(h//2 * R), h//2), cv2.INTER_CUBIC)
new_pic2 = pic[target2_y:target2_y + target2_h, target2_x:target2_x + target2_w]
new_pic2 = cv2.resize(new_pic2, (int(h//2 * R), h//2))
img = np.zeros((h, int(h//2 * R), 3), np.uint8)
img[0:h//2, 0:int(h//2 * R)] = new_pic1
img[h//2:h, 0:int(h//2 * R)] = new_pic2
hmerge = np.hstack((pic, img))
cv2.line(hmerge, (target1_x + target1_w, target1_y), (w, 0), (255, 255, 255), 2)
cv2.line(hmerge, (target2_x + target2_w, target2_y + target2_h), (w, h), (255, 255, 255), 2)
return hmerge
else:
raise ValueError('Make sure the aspect ratio of target is consistent !')
主函数
为方便复现提供了参考的主程序，如果需要批量放大，添加循环即可。
if __name__ == '__main__':
img = cv2.imread(r'./test/Image_11L.jpg')
target1 = [250, 250, 100, 100]
pic, pic1 = Partial_magnification(img, target1, location='lower_right', ratio=4)
target2 = [180, 400, 100, 50]
pic, pic2 = Partial_magnification(pic, target2, location='lower_left', ratio=2)
hmerge = np.hstack((pic1, pic2))
if max(hmerge.shape[0], hmerge.shape[1]) > 1000:
# cv2.namedWindow('merge', 0)
# cv2.resizeWindow('merge', 1000, int(1000*hmerge.shape[0]/hmerge.shape[1]))
# cv2.imshow('merge', hmerge)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
hmerge = cv2.cvtColor(hmerge, cv2.COLOR_BGR2RGB)
fig = plt.figure(figsize=(40, 20)) # figsize 尺寸
plt.imshow(hmerge)
plt.savefig('aaa_bbb.png', dpi=200) # dpi分辨率
plt.show()
else:
plt.imshow(hmerge)
plt.show()
if __name__ == '__main__':
img = cv2.imread(r'./test/Image_11L.jpg')
target1 = [250, 650, 100, 100]
target2 = [450, 400, 100, 100]
pic1 = Partial_magnification_2(img, target1, target2)
hmerge = cv2.cvtColor(pic1, cv2.COLOR_BGR2RGB)
fig = plt.figure(figsize=(30, 20))
fig.patch.set_facecolor('gray')
plt.imshow(hmerge)
plt.savefig('aaa_bbb.png', dpi=300, bbox_inches='tight')
plt.show()
效果
以下是两个函数的测试结果，主程序如上。
总结
以上就是本文的主要内容，都是OpenCV和plt的基本使用上实现图像的局部放大。
致谢
欲尽善本文，因所视短浅，怎奈所书皆是瞽言蒭议。行文至此，诚向予助与余者致以谢意。
