Perceptual Encryption-based Secure Image Retrieval Scheme with Feature Fusion Descriptor
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
With the increasing number of users outsourcing images to cloud servers, secure content-based image retrieval (SCBIR) has become a critical research area. Generating searchable cipher images that provide robustness against attacks and preserve retrieval performance remains a challenge. Towards this, several SCBIR techniques are proposed; however, they often either compromise on search accuracy, lack sufficient security resilience, or suffer from high computational complexity. To address these challenges, we propose a SCBIR scheme that integrates sub-block processing-based perceptual encryption with the feature fusion descriptor. This method enables the calculation of feature histograms in the encrypted domain, striking a better balance between accuracy, security, and computational efficiency. Experimental evaluations conducted on the Coral-1K and GHIM-10K datasets demonstrate that both plain and cipher images exhibit comparable retrieval accuracy. Furthermore, we optimize the proposed scheme by analyzing the impact of varying feature and encryption block sizes, ensuring a comprehensive evaluation of its performance.
Abstract
With the increasing number of users outsourcing images to cloud servers, secure content-based image retrieval (SCBIR) has become a critical research area. Generating searchable cipher images that provide robustness against attacks and preserve retrieval performance remains a challenge. Towards this, several SCBIR techniques are proposed; however, they often either compromise on search accuracy, lack sufficient security resilience, or suffer from high computational complexity. To address these challenges, we propose a SCBIR scheme that integrates sub-block processing-based perceptual encryption with the feature fusion descriptor. This method enables the calculation of feature histograms in the encrypted domain, striking a better balance between accuracy, security, and computational efficiency. Experimental evaluations conducted on the Coral-1K and GHIM-10K datasets demonstrate that both plain and cipher images exhibit comparable retrieval accuracy. Furthermore, we optimize the proposed scheme by analyzing the impact of varying feature and encryption block sizes, ensuring a comprehensive evaluation of its performance.