PixCrypt: Fast Fine-Grained FHE with Range-Aware Caching

A caching-based acceleration mechanism for pixel-level fully homomorphic encryption across CKKS, BFV, and BGV

Chao Wang; Shubing Yang; Xiaoyan Sun; Yan Bai; Jun Dai; Dongfang Zhao· 2026· DOI 10.48550/arXiv.2609.14137

The core problem

Many analytics tasks require secure computation over encrypted data. In particular, fine-grained data such as pixel-level images require higher precision, as every pixel can directly affect outcomes in tasks like tumor segmentation and anomaly detection. Existing privacy-preserving techniques each have limitations: Multi-Party Computation (MPC) is interactive, Differential Privacy (DP) protects only aggregate values, and Partially Homomorphic Encryption (PHE) lacks multiplicative support. None of them can efficiently handle fine-grained data analytics. Fully Homomorphic Encryption (FHE) uniquely enables arbitrary operations on encrypted pixels but remains computationally expensive, posing significant challenges for both software and hardware accelerators. This paper presents PixCrypt, a caching-based acceleration mechanism for fine-grained fully homomorphic encryption. The core insight is to replace expensive fresh ciphertext generation with cache retrieval and coefficient-level operations across CKKS, BFV, and BGV, while randomized reconstruction ensures that ciphertexts do not repeat. The design yields up to 35x faster fine-grained encryption and maintains IND-CPA (Indistinguisha

The authors evaluated PixCrypt on five real-world pixel-level image processing tasks. The experiments demonstrate that PixCrypt achieves up to 35x faster fine-grained encryption compared to baseline FHE encryption. The speedup is attributed to the caching mechanism, which avoids the expensive fresh ciphertext generation for each pixel. The linear noise growth reduces the need for bootstrapping, further improving performance. The NTT load is also lowered, which is particularly beneficial for hardware accelerators. The security analysis confirms that PixCrypt maintains IND-CPA security, ensuring that the encrypted data remains protected. The experiments show that PixCrypt significantly improves the practicality of FHE for privacy-preserving analytics, making it feasible to perform secure computation on fine-grained data such as images.

Many analytics tasks require secure computation over encrypted data. In particular, fine-grained data such as pixel-level images require higher precision, as every pixel can directly affect outcomes in tasks like tumor segmentation and anomaly detection. Existing privacy-preserving techniques each have limitations: Multi-Party Computation (MPC) is interactive, Differential Privacy (DP) protects only aggregate values, and Partially Homomorphic Encryption (PHE) lacks multiplicative support. None of them can efficiently handle fine-grained data analytics. Fully Homomorphic Encryption (FHE) uniquely enables arbitrary operations on encrypted pixels but remains computationally expensive, posing significant challenges for both software and hardware accelerators. This paper presents PixCrypt, a caching-based acceleration mechanism for fine-grained fully homomorphic encryption. The core insight is to replace expensive fresh ciphertext generation with cache retrieval and coefficient-level operations across CKKS, BFV, and BGV, while randomized reconstruction ensures that ciphertexts do not repeat. The design yields up to 35x faster fine-grained encryption and maintains IND-CPA (Indistinguishability under Chosen Plaintext Attack) security.

PixCrypt introduces a range-aware caching mechanism for fine-grained FHE. The key idea is to avoid generating fresh ciphertexts from scratch for every pixel-level operation. Instead, the system maintains a cache of precomputed ciphertexts and performs coefficient-level operations to adapt cached entries to new inputs. This approach is compatible with multiple FHE schemes, including CKKS, BFV, and BGV.

PixCrypt addresses a critical bottleneck in FHE: the high cost of encrypting fine-grained data. By leveraging caching and coefficient-level operations, it achieves substantial speedups without compromising security. The compatibility with multiple FHE schemes (CKKS, BFV, BGV) makes it versatile. The linear noise growth is a key advantage, as it reduces the frequency of bootstrapping, which is a major performance overhead in FHE. The reduced NTT load also benefits hardware implementations. However, the effectiveness of PixCrypt depends on the cache hit rate, which in turn depends on the range of plaintexts and the cache size. Future work could explore adaptive cache management and extension to other FHE schemes. Overall, PixCrypt represents a significant step towards making FHE practical for real-time fine-grained analytics.

CS practitioners and researchers

Opening member content…