286 lines
9.7 KiB
Python
286 lines
9.7 KiB
Python
import logging
|
||
import math
|
||
|
||
import cv2
|
||
import fitz
|
||
import numpy
|
||
from PIL import Image
|
||
|
||
from util import common_util
|
||
|
||
|
||
def capture(image, rectangle):
|
||
"""
|
||
截取图片
|
||
:param image: ndarray
|
||
:param rectangle: 要截取的矩形
|
||
:return: 截取之后的ndarray图片
|
||
"""
|
||
x1, y1, x2, y2 = rectangle
|
||
height, width = image.shape[:2]
|
||
# 确保坐标值在图片范围内
|
||
x1 = max(0, x1)
|
||
y1 = max(0, y1)
|
||
x2 = min(width, x2)
|
||
y2 = min(height, y2)
|
||
return image[int(y1):int(y2), int(x1):int(x2)]
|
||
|
||
|
||
def split(img_path, ratio=1.414, overlap=0.05, x_compensation=3):
|
||
"""
|
||
分割图片
|
||
:param img_path: 图片路径
|
||
:param ratio: 分割后的比例
|
||
:param overlap: 图片之间的覆盖比例
|
||
:param x_compensation: 横向补偿倍率
|
||
:return: 分割后的图片组(NumPy数组形式)
|
||
"""
|
||
split_result = []
|
||
image = cv2.imread(img_path)
|
||
height, width = image.shape[:2]
|
||
hw_ratio = height / width
|
||
wh_ratio = width / height
|
||
|
||
img_name, img_ext = common_util.parse_save_path(img_path)
|
||
if hw_ratio > ratio: # 纵向过长
|
||
new_img_height = width * ratio
|
||
step = width * (ratio - overlap) # 偏移步长
|
||
for i in range(math.ceil(height / step)):
|
||
offset = round(step * i)
|
||
cropped_img = capture(image, [0, offset, width, offset + new_img_height])
|
||
split_path = common_util.get_processed_img_path(f'{img_name}.split_{i}.{img_ext}')
|
||
cv2.imwrite(split_path, cropped_img)
|
||
split_result.append({'img': split_path, 'x_offset': 0, 'y_offset': offset})
|
||
elif wh_ratio > ratio: # 横向过长
|
||
new_img_width = height * ratio
|
||
step = height * (ratio - overlap * x_compensation) # 一般文字是横向的，所以横向截取时增大重叠部分
|
||
for i in range(math.ceil(width / step)):
|
||
offset = round(step * i)
|
||
cropped_img = capture(image, [offset, 0, offset + new_img_width, width])
|
||
split_path = common_util.get_processed_img_path(f'{img_name}.split_{i}.{img_ext}')
|
||
cv2.imwrite(split_path, cropped_img)
|
||
split_result.append({'img': split_path, 'x_offset': offset, 'y_offset': 0})
|
||
else:
|
||
split_result.append({'img': img_path, 'x_offset': 0, 'y_offset': 0})
|
||
return split_result
|
||
|
||
|
||
# def parse_rotation_angles(image):
|
||
# """
|
||
# 判断图片旋转角度，逆时针旋转该角度后为正。可能值['0', '90', '180', '270']
|
||
# :param image: 图片NumPy数组或文件路径
|
||
# :return: 最有可能的两个角度
|
||
# """
|
||
# angles = ['0', '90']
|
||
# model = PaddleClas(model_name='text_image_orientation')
|
||
# clas_result = model.predict(input_data=image)
|
||
# try:
|
||
# clas_result = next(clas_result)[0]
|
||
# if clas_result['scores'][0] < 0.5:
|
||
# return angles
|
||
# angles = clas_result['label_names']
|
||
# except Exception as e:
|
||
# logging.error('获取图片旋转角度失败', exc_info=e)
|
||
# return angles
|
||
|
||
|
||
def rotate(img_path, angle):
|
||
"""
|
||
旋转图片
|
||
:param img_path: 图片NumPy数组
|
||
:param angle: 逆时针旋转角度
|
||
:return: 旋转后的图片NumPy数组
|
||
"""
|
||
if angle == 0:
|
||
return img_path
|
||
image = cv2.imread(img_path)
|
||
height, width = image.shape[:2]
|
||
if angle == 180:
|
||
new_width = width
|
||
new_height = height
|
||
else:
|
||
new_width = height
|
||
new_height = width
|
||
# 绕图像的中心旋转
|
||
# 参数：旋转中心 旋转度数 scale
|
||
matrix = cv2.getRotationMatrix2D((width / 2, height / 2), angle, 1)
|
||
# 旋转后平移
|
||
matrix[0, 2] += (new_width - width) / 2
|
||
matrix[1, 2] += (new_height - height) / 2
|
||
# 参数：原始图像 旋转参数 元素图像宽高
|
||
rotated = cv2.warpAffine(image, matrix, (new_width, new_height))
|
||
|
||
img_name, img_ext = common_util.parse_save_path(img_path)
|
||
rotated_path = common_util.get_processed_img_path(f'{img_name}.rotate_{angle}.{img_ext}')
|
||
cv2.imwrite(rotated_path, rotated)
|
||
return rotated_path
|
||
|
||
|
||
def invert_rotate_point(point, center, angle):
|
||
"""
|
||
反向旋转图片上的点
|
||
:param point: 点
|
||
:param center: 旋转中心
|
||
:param angle: 旋转角度
|
||
:return: 旋转后的点坐标
|
||
"""
|
||
matrix = cv2.getRotationMatrix2D(center, angle, 1)
|
||
if angle != 180:
|
||
# 旋转后平移
|
||
matrix[0, 2] += center[1] - center[0]
|
||
matrix[1, 2] += center[0] - center[1]
|
||
|
||
reverse_matrix = cv2.invertAffineTransform(matrix)
|
||
|
||
point = numpy.array([[point[0]], [point[1]], [1]])
|
||
return numpy.dot(reverse_matrix, point)
|
||
|
||
|
||
def invert_rotate_rectangle(rectangle, center, angle):
|
||
"""
|
||
反向旋转图片上的矩形
|
||
:param rectangle: 矩形
|
||
:param center: 旋转中心
|
||
:param angle: 旋转角度
|
||
:return: 旋转后的矩形坐标
|
||
"""
|
||
if angle == 0:
|
||
return list(rectangle)
|
||
|
||
x1, y1, x2, y2 = rectangle
|
||
|
||
# 计算矩形的四个顶点
|
||
top_left = (x1, y1)
|
||
bot_left = (x1, y2)
|
||
top_right = (x2, y1)
|
||
bot_right = (x2, y2)
|
||
|
||
# 旋转矩形的四个顶点
|
||
rot_top_left = invert_rotate_point(top_left, center, angle).astype(int)
|
||
rot_bot_left = invert_rotate_point(bot_left, center, angle).astype(int)
|
||
rot_bot_right = invert_rotate_point(bot_right, center, angle).astype(int)
|
||
rot_top_right = invert_rotate_point(top_right, center, angle).astype(int)
|
||
|
||
# 找出旋转后矩形的新左上角和右下角坐标
|
||
new_top_left = (min(rot_top_left[0], rot_bot_left[0], rot_bot_right[0], rot_top_right[0]),
|
||
min(rot_top_left[1], rot_bot_left[1], rot_bot_right[1], rot_top_right[1]))
|
||
new_bot_right = (max(rot_top_left[0], rot_bot_left[0], rot_bot_right[0], rot_top_right[0]),
|
||
max(rot_top_left[1], rot_bot_left[1], rot_bot_right[1], rot_top_right[1]))
|
||
|
||
return [new_top_left[0], new_top_left[1], new_bot_right[0], new_bot_right[1]]
|
||
|
||
|
||
def expand_to_a4_size(img_path):
|
||
"""
|
||
以尽量少的方式将图片扩充到a4大小
|
||
:param img_path: 图片路径
|
||
:return: 扩充后的图片NumPy数组和偏移量
|
||
"""
|
||
image = cv2.imread(img_path)
|
||
img_name, img_ext = common_util.parse_save_path(img_path)
|
||
height, width = image.shape[:2]
|
||
x_offset, y_offset = 0, 0
|
||
hw_ratio = height / width
|
||
if hw_ratio >= 1.42:
|
||
exp_w = int(height / 1.414 - width)
|
||
x_offset = int(exp_w / 2)
|
||
exp_img = numpy.zeros((height, x_offset, 3), dtype='uint8')
|
||
exp_img.fill(255)
|
||
image = numpy.hstack([exp_img, image, exp_img])
|
||
elif 1 <= hw_ratio <= 1.40:
|
||
exp_h = int(width * 1.414 - height)
|
||
y_offset = int(exp_h / 2)
|
||
exp_img = numpy.zeros((y_offset, width, 3), dtype='uint8')
|
||
exp_img.fill(255)
|
||
image = numpy.vstack([exp_img, image, exp_img])
|
||
elif 0.72 <= hw_ratio < 1:
|
||
exp_w = int(height * 1.414 - width)
|
||
x_offset = int(exp_w / 2)
|
||
exp_img = numpy.zeros((height, x_offset, 3), dtype='uint8')
|
||
exp_img.fill(255)
|
||
image = numpy.hstack([exp_img, image, exp_img])
|
||
elif hw_ratio <= 0.7:
|
||
exp_h = int(width / 1.414 - height)
|
||
y_offset = int(exp_h / 2)
|
||
exp_img = numpy.zeros((y_offset, width, 3), dtype='uint8')
|
||
exp_img.fill(255)
|
||
image = numpy.vstack([exp_img, image, exp_img])
|
||
else:
|
||
return img_path, 0, 0
|
||
save_path = common_util.get_processed_img_path(f'{img_name}.a4.{img_ext}')
|
||
cv2.imwrite(save_path, image)
|
||
return save_path, x_offset, y_offset
|
||
|
||
|
||
def combined(img1, img2):
|
||
# 获取两张图片的高度和宽度
|
||
height1, width1 = img1.shape[:2]
|
||
height2, width2 = img2.shape[:2]
|
||
|
||
# 确保两张图片的高度相同
|
||
if height1 != height2:
|
||
# 如果高度不同，调整较小高度的图片
|
||
if height1 < height2:
|
||
img1 = cv2.resize(img1, (int(width1 * height2 / height1), height2))
|
||
else:
|
||
img2 = cv2.resize(img2, (int(width2 * height1 / height2), height1))
|
||
|
||
# 再次获取调整后的图片尺寸
|
||
height1, width1 = img1.shape[:2]
|
||
height2, width2 = img2.shape[:2]
|
||
|
||
# 创建一个空白的图像，宽度等于两张图片的宽度之和，高度等于它们共同的高度
|
||
total_width = width1 + width2
|
||
max_height = max(height1, height2)
|
||
combined_img = numpy.zeros((max_height, total_width, 3), dtype=numpy.uint8)
|
||
|
||
# 将img1和img2复制到新的图像中
|
||
combined_img[:height1, :width1] = img1
|
||
combined_img[:height2, width1:width1 + width2] = img2
|
||
return combined_img
|
||
|
||
|
||
def is_photo(img_path):
|
||
"""
|
||
是否是拍照照片
|
||
:param img_path: 图片路径
|
||
:return: True:是照片；False:可能不是照片，也可能在传输过程中相机等信息丢失了
|
||
"""
|
||
img = Image.open(img_path)
|
||
exif = img.getexif()
|
||
if exif:
|
||
# 271:相机制造商, 272:相机型号
|
||
if any(tag in exif for tag in (271, 272)):
|
||
return True
|
||
return False
|
||
|
||
|
||
def pdf_to_imgs(pdf_path, dpi=150):
|
||
pdf_file = None
|
||
# 定义缩放系数（DPI）
|
||
default_dpi = 72
|
||
zoom = dpi / default_dpi
|
||
try:
|
||
# 打开PDF文件
|
||
pdf_file = fitz.open(pdf_path)
|
||
|
||
pdf_imgs = []
|
||
for page in pdf_file:
|
||
# 设置矩阵变换参数
|
||
mat = fitz.Matrix(zoom, zoom)
|
||
# 渲染页面
|
||
pix = page.get_pixmap(matrix=mat)
|
||
# 将渲染结果转换为OpenCV兼容的格式
|
||
img = numpy.frombuffer(pix.samples, dtype=numpy.uint8).reshape((pix.height, pix.width, -1))
|
||
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
||
pdf_imgs.append([img, page.get_text()])
|
||
|
||
return pdf_imgs
|
||
|
||
except Exception as ex:
|
||
logging.getLogger('error').error('pdf转图片失败！', exc_info=ex)
|
||
return None
|
||
finally:
|
||
if pdf_file:
|
||
pdf_file.close()
|