我使用的cv2.imencode比以前快5倍。这是代码

import time
import requests
import base64
import numpy as np
from PIL import Image
from io import BytesIO
from cv2 import imencode


# input: single PIL image
def image_to_base64(image,faster=True):    
    now_time = time.time()
    if faster:        
        image_arr = np.array(image)
        _,byte_data = imencode('.png',image_arr)        
        print('--imencode: ' + str(time.time()-now_time))
    else:
        output_buffer = BytesIO()
        image.save(output_buffer,format='PNG')
        byte_data = output_buffer.getvalue()
        print('--image.save:' + str(time.time()-now_time))

    now_time = time.time()
    encoded_input_string  = base64.b64encode(byte_data)
    print('--base64.b64encode: ' + str(time.time()-now_time))

    now_time = time.time()
    input_string = encoded_input_string.decode("utf-8")
    print('--encoded_input_string.decode: ' + str(time.time()-now_time))  

    return input_string

img_url = "https://www.cityscapes-dataset.com/wordpress/wp-content/uploads/2015/07/stuttgart03.png"
response = requests.get(img_url)
img = Image.open(BytesIO(response.content))
now_time = time.time()
input_string = image_to_base64(img,faster=True)
print('total: ' + str(time.time()-now_time))

我想知道是否有可以更快运行的解决方案。

根据评论中的建议，我尝试了pyvips，如下所示：

#!/usr/bin/env python3
import requests
import base64
import numpy as np
from PIL import Image
from io import BytesIO
from cv2 import imencode
import pyvips

def vips_2PNG(image,compression=6):
    # Convert PIL Image to Numpy array
    na = np.array(image)
    height,width,bands = na.shape

    # Convert Numpy array to Vips image
    dtype_to_format = {
       'uint8': 'uchar','int8': 'char','uint16': 'ushort','int16': 'short','uint32': 'uint','int32': 'int','float32': 'float','float64': 'double','complex64': 'complex','complex128': 'dpcomplex',}
    linear = na.reshape(width * height * bands)
    vi = pyvips.Image.new_from_memory(linear.data,height,bands,dtype_to_format[str(na.dtype)])

    # Save to memory buffer as PNG
    data = vi.write_to_buffer(f".png[compression={compression}]")
    return data

def vips_including_reading_from_disk(image):
    # Load image from disk
    image = pyvips.Image.new_from_file('stuttgart.png',access='sequential')
    # Save to memory buffer as PNG
    data = image.write_to_buffer('.png')
    return data

def faster(image):
    image_arr = np.array(image)
    _,image_arr)        
    return byte_data

def orig(image,faster=True):    
    output_buffer = BytesIO()
    image.save(output_buffer,format='PNG')
    byte_data = output_buffer.getvalue()
    return byte_data

# img_url = "https://www.cityscapes-dataset.com/wordpress/wp-content/uploads/2015/07/stuttgart03.png"
filename = 'stuttgart.png'
img = Image.open(filename)

# r = orig(img)
# print(len(r))
# %timeit r = orig(img)

# r = faster(img)
# print(len(r))
# %timeit r = faster(img)

# r = vips_including_reading_from_disk(filename)
# print(len(r))
# %timeit r = vips_including_reading_from_disk(filename)

# r = vips_2PNG(img,0)
# print(len(r))
# %timeit r = vips_2PNG(img,0)

我当时正在考虑在文件大小和速度之间权衡compression参数。这就是我得到的-我不会比较绝对值，而是查看机器上彼此之间的性能 relative ：

               Filesize        Time
PIL            1.7MB           1.12s
OpenCV         2.0MB           173ms   <--- COMPARE
vips(comp=0)   6.2MB           66ms
vips(comp=1)   2.0MB           132ms   <--- COMPARE
vips(comp=2)   2.0MB           153ms

我将箭头放在要比较的箭头旁边。