[Bug] Transparent images with white fonts are not handled correctly.

[SWHL]

请提供下述完整信息以便快速定位问题
(Please provide the following information to quickly locate the problem)

• 系统环境/System Environment: macOS

• 使用的是哪门语言的程序/Which programing language: Python

• 所使用语言相关版本信息/Version: 3.10.13

• OnnxRuntime版本/OnnxRuntime Version: 1.17.0

• 使用当前库的版本/Use version: rapidocr_onnxruntime==1.3.14

• 可复现问题的demo和文件/Demo of reproducible problems:

  import cv2
  import numpy as np
  
  
  def cvt_four_to_three(img: np.ndarray) -> np.ndarray:
      """RGBA → BGR"""
      r, g, b, a = cv2.split(img)
      new_img = cv2.merge((b, g, r))
  
      not_a = cv2.bitwise_not(a)
      not_a = cv2.cvtColor(not_a, cv2.COLOR_GRAY2BGR)
  
      new_img = cv2.bitwise_and(new_img, new_img, mask=a)
      new_img = cv2.add(new_img, not_a)
      return new_img
  
  
  img_path = "1.png"
  img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)
  res_img = cvt_four_to_three(img)
  
  cv2.imwrite("res.png", res_img)
  print("ok")

• 所用图像/ Use image：
  1.png.zip

• 程序执行结果（注意下面有一张图，只不过是全白的）：
  (Result (note that there is an image below, but it is completely white))
  

[SWHL]

Maybe a solution:

def cvt_four_to_three(img: np.ndarray) -> np.ndarray:
    """RGBA → BGR"""
    r, g, b, a = cv2.split(img)
    new_img = cv2.merge((b, g, r))

    not_a = cv2.bitwise_not(a)
    not_a = cv2.cvtColor(not_a, cv2.COLOR_GRAY2BGR)

    new_img = cv2.bitwise_and(new_img, new_img, mask=a)

    mean_color = np.mean(new_img)
    if mean_color <= 0.0:
        new_img = cv2.add(new_img, not_a)
    else:
        new_img = cv2.bitwise_not(new_img)
    return new_img

[SWHL]

The issue has been fixed in rapidocr_onnxruntime>=1.3.15, rapidocr_openvino>=1.3.15 and rapidocr_paddle>=1.3.17.