allow for higher-precision pHash computation

[tgpfeiffer]

This is an attempt to fix #30

(Disclaimer: I implemented this using some support from Copilot.)

This goes the straightforward way and replaces u64 in the ImageHash struct by a Vec<u8>. I am not sure about the performance implications as I don't know the numbers before this PR, but on my machine the benchmarks now look like

$ python compare.py
Found 20 images in ../../../imgs/test/. Running benchmarks for 100 runs...

Benchmarking dHash...

Benchmark Results for dHash (in seconds per image):

Metric      imgdd (ns)  imagehash (ns)
----------------------------------------------
min_time    1316625.500000 3898336.000000
max_time    2176291.550000 4588001.600000
avg_time    1470818.250000 3979595.414500
median_time 1410523.850000 3951972.775000

Percentage Difference (imgdd vs. imagehash):

Metric      Difference (%) 
----------------------------
min_time    66.23          
max_time    52.57          
avg_time    63.04          
median_time 64.31          


Benchmarking aHash...

Benchmark Results for aHash (in seconds per image):

Metric      imgdd (ns)  imagehash (ns)
----------------------------------------------
min_time    1296290.500000 3864925.150000
max_time    2131858.950000 4151851.300000
avg_time    1462870.159000 3938232.084500
median_time 1409924.325000 3938366.125000

Percentage Difference (imgdd vs. imagehash):

Metric      Difference (%) 
----------------------------
min_time    66.46          
max_time    48.65          
avg_time    62.85          
median_time 64.20          


Benchmarking pHash...

Benchmark Results for pHash (in seconds per image):

Metric      imgdd (ns)  imagehash (ns)
----------------------------------------------
min_time    1340875.600000 4332915.000000
max_time    1787807.250000 4927223.450000
avg_time    1455338.274500 4425243.710500
median_time 1418410.925000 4392500.725000

Percentage Difference (imgdd vs. imagehash):

Metric      Difference (%) 
----------------------------
min_time    69.05          
max_time    63.72          
avg_time    67.11          
median_time 67.71          


Benchmarking pHash256...

Benchmark Results for pHash256 (in seconds per image):

Metric      imgdd (ns)  imagehash (ns)
----------------------------------------------
min_time    1337245.050000 4625274.400000
max_time    1836423.200000 6072616.000000
avg_time    1479914.261000 4774408.488000
median_time 1429946.300000 4707407.375000

Percentage Difference (imgdd vs. imagehash):

Metric      Difference (%) 
----------------------------
min_time    71.09          
max_time    69.76          
avg_time    69.00          
median_time 69.62

...

I've opted for the smallest possible change to the code, which made the external interface a bit ugly: I added an optional hash_size parameter to all the exposed functions, which is ignored by all algorithms except pHash. I guess a better way would be to define an enum where each variant corresponds to one algorithm and contains the parameters for this algorithm, but it changes the interface in a backwards-incompatible way and requires more work on the Python/Rust border, so I wanted to get your feedback first. Please let me know what you think @aastopher.

[deepsource-io]

Here's the code health analysis summary for commits 3dcfca5..a47380b. View details on DeepSource ↗.

Analysis Summary

Analyzer	Status	Summary	Link
Rust	✅ Success		View Check ↗

---

💡 If you’re a repository administrator, you can configure the quality gates from the settings.