Optimization encoding in 124 times

[homm]

Before:

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').copy()
   ...: %time im.size, encode(im.copy(), 9, 9)
CPU times: user 12.1 s, sys: 12.1 ms, total: 12.1 s
Wall time: 12.1 s
Out[1]: 
((2000, 1334),
 '|7E-z7oyM{8xM{wKTdWFNuwdMepHrE%LV[kVV@jFOVV@BBS$r@Naen={rrR7OrRQNaMKXmw|?aRPaLt6w0x[RQkqa1xuD*xZNuWEt6n%s-Vtf+M|x[ixxswctQnPNGR-RkXma0xtRkS~jF$%WXI;S~M|X8RjxYniR+ozs-')

After:

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').copy()
   ...: %time im.size, encode(im, 9, 9)
CPU times: user 141 ms, sys: 2.01 ms, total: 143 ms
Wall time: 143 ms
Out[1]: 
((2000, 1334),
 '|7E-z7oyM{8xM{wKTdWFNuwdMepHrE%LV[kVV@jFOVV@BBS$r@Naen={rrR7OrRQNaMKXmw|?aRPaLt6w0x[RQkqa1xuD*xZNuWEt6n%s-Vtf+M|x[ixxswctQnPNGR-RkXma0xtRkS~jF$%WXI;S~M|X8RjxYniR+ozs-')

Also improved:

Pillow image is only closed if it was opened inside the encoder.

Update

Improved more:

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').copy()
   ...: %time im.size, encode(im, 9, 9)
CPU times: user 70.4 ms, sys: 1.01 ms, total: 71.4 ms
Wall time: 71.4 ms
Out[1]: 
((2000, 1334),
 '|7E-z7oyM{8xM{wKTdWFNuwdMepHrE%LV[kVV@jFOVV@BBS$r@Naen={rrR7OrRQNaMKXmw|?aRPaLt6w0x[RQkqa1xuD*xZNuWEt6n%s-Vtf+M|x[ixxswctQnPNGR-RkXma0xtRkS~jF$%WXI;S~M|X8RjxYniR+ozs-')

[homm]

For normal-size images speedup is also about 33x 53x 129x.

Before:

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').resize((360, 240), Image.BICUBIC)
   ...: assert encode(im.copy(), 6, 4) == 'W7EUfLs:M{8yM{wJ$%MepHrE%LV[NaVtBUXSsBNbVuOrM|NaMeXm'
   ...: %timeit encode(im.copy(), 6, 4)
125 ms ± 236 μs per loop (mean ± std. dev. of 7 runs, 10 loops each)

After:

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').resize((360, 240), Image.BICUBIC)
   ...: assert encode(im, 6, 4) == 'W7EUfLs:M{8yM{wJ$%MepHrE%LV[NaVtBUXSsBNbVuOrM|NaMeXm'
   ...: %timeit encode(im, 6, 4)
1.44 ms ± 3.13 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Update

In [1]: from blurhash import encode, Image
   ...: im = Image.open('/Code/imgs/bologna2k.jpg').resize((360, 240), Image.BICUBIC)
   ...: assert encode(im, 6, 4) == 'W7EUfLs:M{8yM{wJ$%MepHrE%LV[NaVtBUXSsBNbVuOrM|NaMeXm'
   ...: %timeit encode(im, 6, 4)
971 μs ± 4.82 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

[homm]

Also back ported to upstream: woltapp/blurhash#256

[homm]

I have updated branch to match the latest version from woltapp/blurhash#256

@lautat Are you interested in this improvements?

[vellnes]

@lautat please note that
We will be very grateful for the optimization of the algorithm