python3 通过头文件判断文件类型

CODE_TYPE_DIC = {
    '68746D6C3E': 'html',
    'd0cf11e0a1b11ae10000': 'xls',
    '44656C69766572792D64': 'eml',
    'ffd8ffe000104a464946': 'jpg',
    '89504e470d0a1a0a0000': 'png',
    '47494638396126026f01': 'gif',
    '49492a00227105008037': 'tif',
    '424d228c010000000000': 'bmp',
    '424d8240090000000000': 'bmp',
    '424d8e1b030000000000': 'bmp',
    '41433130313500000000': 'dwg',
    '3c21444f435459504520': 'html',
    '3c21646f637479706520': 'htm',
    '48544d4c207b0d0a0942': 'css',
    '696b2e71623d696b2e71': 'js',
    '7b5c727466315c616e73': 'rtf',
    '38425053000100000000': 'psd',
    '46726f6d3a203d3f6762': 'eml',
    'd0cf11e0a1b11ae10000': 'doc',
    'd0cf11e0a1b11ae10000': 'vsd',
    '5374616E64617264204A': 'mdb',
    '252150532D41646F6265': 'ps',
    '255044462d312e350d0a': 'pdf',
    '2e524d46000000120001': 'rmvb',
    '464c5601050000000900': 'flv',
    '00000020667479706d70': 'mp4',
    '49443303000000002176': 'mp3',
    '000001ba210001000180': 'mpg',
    '3026b2758e66cf11a6d9': 'wmv',
    '52494646e27807005741': 'wav',
    '52494646d07d60074156': 'avi',
    '4d546864000000060001': 'mid',
    '504b0304140000080044': 'zip',
    '504b03040a0000080000': 'zip',
    '504b03040a0000000000': 'zip',
    '526172211a0700cf9073': 'rar',
    '235468697320636f6e66': 'ini',
    '504b03040a0000000000': 'jar',
    '4d5a9000030000000400': 'exe',
    '3c25402070616765206c': 'jsp',
    '4d616e69666573742d56': 'mf',
    '3c3f786d6c2076657273': 'xml',
    '494e5345525420494e54': 'sql',
    '7061636b616765207765': 'java',
    '406563686f206f66660d': 'bat',
    '1f8b0800000000000000': 'gz',
    '6c6f67346a2e726f6f74': 'properties',
    'cafebabe0000002e0041': 'class',
    '49545346030000006000': 'chm',
    '04000000010000001300': 'mxp',
    '504b0304140006000800': 'docx',
    'd0cf11e0a1b11ae10000': 'wps',
    '6431303a637265617465': 'torrent',
}


# 字节码转16进制字符串
def bytes2hex(b):
    num = len(b)
    hex_str = u''
    for i in range(num):
        t = u'%x' % b[i]
        if len(t) % 2:
            hex_str += u'0'
        hex_str += t
    return hex_str.upper()


# 获取文件类型
def file_type(filename, n=20):
    with open(filename, 'rb') as f:
        bins = f.read(n)
    bins = bytes2hex(bins)  # 转码
    bins = bins.lower()  # 小写
    f_type = 'unknown'
    for hcode in CODE_TYPE_DIC.keys():
        lens = len(hcode)  # 需要的长度
        if bins[0:lens] == hcode:
            f_type = CODE_TYPE_DIC[hcode]
            break
    if f_type == 'unknown':  # 全码未找到,优化处理,码表取5位验证
        bins = bins[0:5]
        for hcode in CODE_TYPE_DIC.keys():
            if len(hcode) > 5 and bins == hcode[0:5]:
                f_type = CODE_TYPE_DIC[hcode]
                break
    return f_type


if __name__ == '__main__':
    print(file_type('01.jpg'))

参考链接:https://blog.csdn.net/privateobject/article/details/78069500

posted @ 2020-10-28 15:59  黄土地上的黑石头  阅读(593)  评论(0编辑  收藏  举报