Human-Centred Encrypted JPEG Retrieval for Substation Inspection: A Framework and an Evaluation Protocol
Substation inspection produces a rapidly growing volume of images that must be stored, searched and reused, yet these images disclose sensitive grid asset information and therefore cannot be handed to an untrusted cloud in plaintext. Content-based image retrieval (CBIR) over encrypted JPEG images reconciles this conflict, but existing schemes are evaluated almost exclusively by retrieval accuracy and file-size overhead; whether an inspector can actually use, understand and trust such a system has rarely been examined. This paper presents a human-centred encrypted JPEG retrieval framework for intelligent substations. The image owner encrypts each image by stream cipher and block scrambling in a format-compatible manner, so that the JPEG bitstream remains compressible and the file size is not expanded. The cloud server extracts the variable-length integer (VLI) code lengths from the cipher-images and learns a fixed-length representation with a convolutional neural network; similarity is measured by Euclidean distance and the top-k cipher-images are returned to the operator. We report the encryption security under the chosen-plaintext and known-background-knowledge models, the retrieval accuracy and the computational cost, and we complement them with an evaluation protocol that specifies how the usability and trust consequences of encryption are to be measured with inspection personnel, namely task completion time, perceived workload (NASA-TLX), usability (SUS) and trust in the system, together with a calibration index that relates reported trust to observed retrieval precision. No operator data are reported here: the protocol is stated in advance of data collection so that it can be reviewed and replicated. On this basis we discuss the privacy–utility–human trade-off and the deployment constraints of power-system compliance, and we argue that operator-centred evidence should complement accuracy-oriented benchmarks for encrypted retrieval.
Authors
- Xu Yang (ORCID: https://orcid.org/0000-0003-0553-4581)
- Shilei Ding
- Kun Hu (ORCID: https://orcid.org/0000-0001-5044-7327)
- Hongyu Zhou (ORCID: https://orcid.org/0000-0003-4282-8220)
- Xia Guo
- Gang Wang
- Bin Li
- Hui Wang
- Da Fang
Institutions
- State Grid Corporation of China (China) (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-11570767/v1
- Primary Topic
- Image Retrieval and Classification Techniques
- Type
- preprint