James Lin James Lin - 20 days ago 7
Python Question

Python image library (PIL), how to compress image into desired file size?

I got a requirement to compress any uploaded images less than 500kb in file size, I have searched on google and all I can see is:

>>> foo = foo.resize((160,300),Image.ANTIALIAS)
>>> foo.save("path\\to\\save\\image_scaled.jpg",quality=95)


If I go with this approach I will have to check if the image is less than 500kb after compress, if not then go for lower quality and size.

Is there a better way to do it?

Answer

JPEG compression is not predictable beforehand. The method you described, compress & measure & try again, is the only way I know.

You can try compressing a number of typical images with different quality settings to get an idea of the optimum starting point, plus a way of guessing how changes to the setting will affect the size. That will get you to zero in on the optimum size without too many iterations.

You can also pass a file-like object to the save function that doesn't bother to write to disk, just counts the bytes. Once you've determined the best settings then you can save it again to an actual file.

Edit: Here's an implementation of a suitable byte counting file object. Just check size after the save.

class file_counter(object):
    def __init__(self):
        self.position = self.size = 0

    def seek(self, offset, whence=0):
        if whence == 1:
            offset += self.position
        elif whence == 2:
            offset += self.size
        self.position = min(offset, self.size)
        self.size = max(self.size, self.position)

    def tell(self):
        return self.position

    def write(self, string):
        self.position += len(string)
        self.size = max(self.size, self.position)

Edit 2: Here's a binary search using the above to get the optimal quality in the smallest number of attempts.

def smaller_than(im, size, guess=70, subsampling=1, low=1, high=100):
    while low < high:
        counter = file_counter()
        im.save(counter, format='JPEG', subsampling=subsampling, quality=guess)
        if counter.size < size:
            low = guess
        else:
            high = guess - 1
        guess = (low + high + 1) // 2
    return low
Comments