Encrypting and operating with 2D numpy arrays (images)

[elyasgoli]

Description
When I try Demo_3_Float_CKKS.py with a 2-dimensional numpy array, I get an error.

Code To Reproduce Error

arr_x [[1. 2. 3.]
 [4. 5. 6.]
 [7. 8. 9.]]
Traceback (most recent call last):
  File "/home/elyas/Pyfhel/examples/Try_Float_CKKS.py", line 60, in <module>
    ptxt_x = HE.encodeFrac(arr_x)   # Creates a PyPtxt plaintext with the encoded arr_x
  File "Pyfhel/Pyfhel.pyx", line 742, in Pyfhel.Pyfhel.Pyfhel.encodeFrac
ValueError: Buffer has wrong number of dimensions (expected 1, got 2)

Setup:

• OS: Ubuntu
• Python: 3.9.2
• C compiler version: GCC 10.2.1 20210110
• Pyfhel Version: most-recent

[ShokofehVS]

Hi @elyasgoli, I just want to add that there is a nice example of how encrypting and performing simple arithmetic operations over 2-D matrices in this issue.

[elyasgoli]

Hi Shokofeh, Thank you! That's helpful! My goal is actually to encrypt images with CKKS or BFV. For that purpose, I need to encrypt a Numpy 2D array as a whole, not by breaking it down row by row. Is it possible to do that with PyFHEL? Many thanks, Elyas

…

On Fri, Sep 1, 2023 at 4:55 AM Shokofeh VahidianSadegh < ***@***.***> wrote: Hi @elyasgoli <https://github.com/elyasgoli>, I just want to add that there is a nice example of how encrypting and performing simple arithmetic operations over 2-D matrices in this issue <#183 (comment)>. — Reply to this email directly, view it on GitHub <#212 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A7HSUU7XPBOUCHB5NIO235DXYHEL7ANCNFSM6AAAAAA4HAH56U> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- Elyas Goli, Ph.D. Principal Machine Learning Engineer Peak Artificial Intelligence Cell Phone: 217-904-4873 Email: ***@***.***

[AlexanderViand-Intel]

You virtually always want to flatten images before encryption, as FHE ciphertexts are effectively vectors and FHE offers SIMD operations over vectors. For advanced use cases, more complex patterns might be beneficial, but expressing your FHE computation as a vectorized computation is a good start for most applications.