A privacy-preserving scheme for indoor localization was proposed to tackle the challenge of securing user location information in cloud-based multi-map environments. Bloom filters were employed to encode the distinctive fingerprint features of each map, thereby enabling fast and efficient map-level localization through inner product computations. For fine-grained localization in the ciphertext domain, an improved k-nearest neighbors algorithm (KNN) based on the modified Sorensen distance was integrated with Paillier homomorphic encryption. To further protect fingerprint data and resist inference attacks, a fingerprint reconstruction mechanism was introduced that permuted the fingerprint information during localization queries, breaking the arrangement rules of the original signals, thereby ensuring enhanced protection of both user privacy and server-side data confidentiality. Experimental evaluation on real-world datasets demonstrated that the proposed method consistently maintained high localization accuracy and low latency while keeping computational and communication overhead at a practical level. Security analysis confirmed that user location privacy was preserved throughout the entire process and that server-side data remained well protected against potential leakage.