Reapply "优化照片涂抹功能架构"

This reverts commit 13575fb2aa.

util/image_util.py

@@ -0,0 +1,202 @@
+import logging
+import math
+import urllib.request
+
+import cv2
+import numpy
+from paddleclas import PaddleClas
+
+
+def read(image_path):
+    """
+    从网络或本地读取图片
+    :param image_path: 网络或本地路径
+    :return: NumPy数组形式的图片
+    """
+    if image_path.startswith("http"):
+        # 发送HTTP请求并获取图像数据
+        resp = urllib.request.urlopen(image_path)
+        # 将数据读取为字节流
+        image_data = resp.read()
+        # 将字节流转换为NumPy数组
+        image_np = numpy.frombuffer(image_data, numpy.uint8)
+        # 解码NumPy数组为OpenCV图像格式
+        image = cv2.imdecode(image_np, cv2.IMREAD_COLOR)
+    else:
+        image = cv2.imread(image_path)
+    return image
+
+
+def capture(image, rectangle):
+    """
+    截取图片
+    :param image: 图片NumPy数组
+    :param rectangle: 要截取的矩形
+    :return: 截取之后的图片NumPy
+    """
+    x1, y1, x2, y2 = rectangle
+    return image[int(y1):int(y2), int(x1):int(x2)]
+
+
+def split(image, ratio=1.414, overlap=0.05):
+    """
+    分割图片，只分割过长的图片，暂不处理过宽的图片
+    :param image:图片，可以是NumPy数组或文件路径
+    :param ratio: 分割后的比例
+    :param overlap: 图片之间的覆盖比例
+    :return: 分割后的图片组(NumPy数组形式)
+    """
+    split_result = []
+    if isinstance(image, str):
+        image = read(image)
+    # 获取图片的宽度和高度
+    height, width = image.shape[:2]
+    # 计算宽高比
+    img_ratio = height / width
+    # 检查是否需要裁剪
+    if img_ratio > ratio:
+        split_ratio = ratio - overlap
+        # 分割后的高度
+        new_img_height = width * ratio
+        for i in range(math.ceil(height / (width * split_ratio))):
+            offset = round(width * split_ratio * i)
+            # 参数形式为[y1:y2, x1:x2]
+            cropped_img = capture(image, [0, offset, width, offset + new_img_height])
+            split_result.append({"img": cropped_img, "x_offset": 0, "y_offset": offset})
+    else:
+        split_result.append({"img": image, "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(image, angle):
+    """
+    旋转图片
+    :param image: 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 image
+    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))
+    return rotated
+
+
+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(image, center=False):
+    """
+    将图片扩充到a4大小
+    :param image: 图片NumPy数组
+    :param center: 是否将原图置于中间
+    :return: 扩充后的图片NumPy数组和偏移量
+    """
+    h, w = image.shape[:2]
+    offset_x, offset_y = 0, 0
+    if h * 1.0 / w >= 1.42:
+        exp_w = int(h / 1.414 - w)
+        if center:
+            offset_x = int(exp_w / 2)
+            exp_img = numpy.zeros((h, offset_x, 3), dtype="uint8")
+            exp_img.fill(255)
+            image = numpy.hstack([exp_img, image, exp_img])
+        else:
+            exp_img = numpy.zeros((h, exp_w, 3), dtype="uint8")
+            exp_img.fill(255)
+            image = numpy.hstack([image, exp_img])
+    elif h * 1.0 / w <= 1.40:
+        exp_h = int(w * 1.414 - h)
+        if center:
+            offset_y = int(exp_h / 2)
+            exp_img = numpy.zeros((offset_y, w, 3), dtype="uint8")
+            exp_img.fill(255)
+            image = numpy.vstack([exp_img, image, exp_img])
+        else:
+            exp_img = numpy.zeros((exp_h, w, 3), dtype="uint8")
+            exp_img.fill(255)
+            image = numpy.vstack([image, exp_img])
+    return image, offset_x, offset_y