opencv无法正确读取png图片导致ocr识别失败

[donggoing]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

#16104 和 #16099 (comment) 中的第一张图在读取后文字完全不可见，导致文字识别失败

运行Demo保存图片后如下：

🏃‍♂️ Environment (运行环境)

macOS m4 & linux

paddleocr 3.1.0
paddlepaddle 3.1.0
paddlex 3.1.3

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

import cv2
from paddlex.inference.common.reader.image_reader import ReadImage
img = ReadImage().read("sprite_7d0cc3c385.png")
cv2.imwrite("sprite_7d0cc3c385_1.png", img)

[zhangyubo0722]

收到，感谢反馈，这个问题我们尽快排查

[jingsongliujing]

这个问题应该是使用image_reader读取的图片包含 alpha 通道而没有正确处理导致的文字丢失，估计得修复paddlex这边，或者建议您这边：https://github.com/PaddlePaddle/PaddleX/blob/release/3.1/paddlex/inference/common/reader/image_reader.py
或者使用临时方案修复一下，针对这种图片：

import cv2
import numpy as np
from paddlex.inference.common.reader.image_reader import ReadImage

# 读取图片
img = ReadImage().read("sprite_7d0cc3c385.png")

# 处理alpha通道（透明背景）
if img.ndim == 3 and img.shape[2] == 4:
    # 分离通道
    b, g, r, a = cv2.split(img)
    # 创建白色背景
    white = np.ones_like(b) * 255
    # 透明区域替换为白色
    b = cv2.bitwise_and(b, a) + cv2.bitwise_and(white, ~a)
    g = cv2.bitwise_and(g, a) + cv2.bitwise_and(white, ~a)
    r = cv2.bitwise_and(r, a) + cv2.bitwise_and(white, ~a)
    # 合并为3通道图片
    img = cv2.merge([b, g, r])
elif img.shape[-1] == 3 and ReadImage().format == "RGB":
    img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

# 保存处理后的图片
cv2.imwrite("sprite_7d0cc3c385_1.png", img)