Python exifread 读取修改过的 jpeg 信息错误问题修复

本文最后更新于：2025年12月8日下午

有时需要修改 JPEG 图像中的 exif 信息，但是修改过的信息直接用 exifread 库读取报错，本文复现问题并记录解决方案。

简介

大疆的无人机拍摄的 jpeg 图像中自带复杂丰富的 exif 和 xmp 信息，我们可以用 python 进行读取，也可以编辑，但是编辑过后 exifread 库无法顺利读取。

问题复现

原始图像：https://uipv4.zywvvd.com:33030/HexoFiles/demo_files/ori.JPG

此时是可以用 exifread 库读取信息的：

import exifread

image_path =  'ori.JPG'
with open(image_path, 'rb') as fileobj:
    tags = exifread.process_file(fileobj, details=False)
    print(tags)

修改 xmp 信息：

from pyexiv2 import Image as xmpImage
from pyexiv2 import ImageData
import vvdutils as vv

# def phase_all_exif_info(image_path):
#     img = xmpImage(image_path)
#     total_info = dict()

#     xmp_data = img.read_xmp()
#     total_info.update(xmp_data)

#     exif_info = img.read_exif()
#     total_info.update(exif_info)

#     return total_info

def signed_str_to_float(s):
    try:
        # 去除前后空格后直接转换
        return float(s.strip())
    except ValueError as e:
        print(f"转换失败: 输入 '{s}' 不是有效的浮点数字符串")
        return None


def phase_all_exif_info(image_path):
    img = xmpImage(image_path)
    total_info = dict()

    xmp_data = img.read_xmp()
    total_info.update(xmp_data)

    exif_info = img.read_exif()
    total_info.update(exif_info)

    return total_info


'''
修改图像中EXIF和XMP数据, 原地操作，直接修改存储到当前图像
数据示例如下：
changes = { 'Xmp.drone-dji.AbsoluteAltitude': '+10000' }
'''  
def modify_xmp_and_exif(image_path, xmp_changes=None, exif_changes=None):
    """
    只修改 XMP 元数据，不操作像素数据
    """
    try:
        # 1. 打开图片读取元数据（仅加载元数据，不加载完整像素）
        with xmpImage(image_path) as img:
            # 如果提供了 EXIF 修改，应用它
            if exif_changes:
                img.modify_exif(exif_changes)
            else:
                img.modify_exif({})
            
            # 如果提供了 XMP 修改，应用它
            if xmp_changes:
                img.modify_xmp(xmp_changes)
            
            # 3. 获取修改后的完整元数据（包括 EXIF、XMP）
            exif = img.read_exif()
            xmp = img.read_xmp()
        
        # 4. 用原始图片创建 ImageData 对象（承载像素和原始元数据）
        with open(image_path, 'rb') as f:
            img_data = ImageData(f.read())
        
        # 5. 将修改后的元数据写入 ImageData（仅更新元数据区）
        img_data.modify_exif(exif)
        img_data.modify_xmp(xmp)

    except FileNotFoundError:
        print(f"错误：找不到文件 '{image_path}'")
    except Exception as e:
        print(f"处理图片时发生错误: {e}")


if __name__ == '__main__':
    utm_zone = "WGS84 UTM 45N"
    ariport_offset_lon =  1000  # cur_pos - airport_pos
    ariport_offset_lat =  1000 
    ariport_offset_alt = 1000

    image_file = 'ori.JPG'

    exif_info = phase_all_exif_info(image_file)

    lon = exif_info['Xmp.drone-dji.GpsLongitude']
    lon = signed_str_to_float(lon)
    lat = exif_info['Xmp.drone-dji.GpsLatitude']
    lat = signed_str_to_float(lat)
    alt = exif_info['Xmp.drone-dji.AbsoluteAltitude']
    alt = signed_str_to_float(alt)
    pt = [lon, lat]
    point_image = vv.Point(pt[1], pt[0], utm_zone=utm_zone)
    print(f'1:lon:{lon}, lat:{lat}')
    point_image_x_utm = point_image.x_utm - ariport_offset_lon
    point_image_y_utm = point_image.y_utm - ariport_offset_lat
    new_point = vv.Point.from_xy_utm(utm_zone, point_image_x_utm, point_image_y_utm)

    print(f'2:lon:{new_point.lon}, lat:{new_point.lat}')

    update_lon = f'+{new_point.lon}' if new_point.lon > 0  else f'-{new_point.lon}'
    update_lat = f'+{new_point.lat}' if new_point.lat > 0  else f'-{new_point.lat}'
    changes = {
        'Xmp.drone-dji.AbsoluteAltitude': '+10000',
        'Xmp.drone-dji.GpsLongitude': update_lon,
        'Xmp.drone-dji.GpsLatitude': update_lat 
    }
    modify_xmp_and_exif(image_file, changes)
    out_exif_info = phase_all_exif_info(image_file)
    pass

修改前后：

1 2	`1:lon:87.616024029, lat:44.600398776 2:lon:87.603330963344, lat:44.591463976635`

可以在图像中看到修改成功。

但是此时用 exifread 执行相同代码却只能获取空字典。

问题定位

在 exifread 库的 jpeg.py 中的 find_jpeg_exif 函数里：


def find_jpeg_exif(fh: BinaryIO, data, fake_exif) -> tuple:
    logger.debug("JPEG format recognized data[0:2]=0x%X%X", ord_(data[0]), ord_(data[1]))

    base, fake_exif = _get_initial_base(fh, data, fake_exif)

    # Big ugly patch to deal with APP2 (or other) data coming before APP1
    fh.seek(0)
    # in theory, this could be insufficient since 64K is the maximum size--gd
    data = fh.read(base + 4000)

    base = _get_base(base, data)

    fh.seek(base + 12)
    if ord_(data[2 + base]) == 0xFF and data[6 + base : 10 + base] == b"Exif":
        # detected EXIF header
        offset = fh.tell()
        endian = fh.read(1)
        # HACK TEST:  endian = 'M'
    elif ord_(data[2 + base]) == 0xFF and data[6 + base : 10 + base + 1] == b"Ducky":
        # detected Ducky header.
        logger.debug(
            "EXIF-like header (normally 0xFF and code): 0x%X and %s",
            ord_(data[2 + base]),
            data[6 + base : 10 + base + 1],
        )
        offset = fh.tell()
        endian = fh.read(1)
    elif ord_(data[2 + base]) == 0xFF and data[6 + base : 10 + base + 1] == b"Adobe":
        # detected APP14 (Adobe)
        logger.debug(
            "EXIF-like header (normally 0xFF and code): 0x%X and %s",
            ord_(data[2 + base]),
            data[6 + base : 10 + base + 1],
        )
        offset = fh.tell()
        endian = fh.read(1)
    else:
        # no EXIF information
        msg = "No EXIF header expected data[2+base]==0xFF and data[6+base:10+base]===Exif (or Duck)"
        msg += "Did get 0x%X and %s" % (ord_(data[2 + base]), data[6 + base : 10 + base + 1])
        raise InvalidExif(msg)
    return offset, endian, fake_exif