A content-based image retrieval system for visually protected image databases

This paper presents a content-based image retrieval mechanism for visually protected image databases. Visual protection is provided as a means to prevent illegal use of a database by those who could break access control, such as an untrustworthy server provider or intruders. Considering the widespread use of the JPEG standard, we focus on JPEG-coded images in the current work and exploit several advantages of the Discrete Cosine Transform (DCT). The visually protected images were obtained by applying DCT-based scrambling to the original images. The proposed scheme directly extracts a DC image from the lowest frequency DCT coefficients of a visually protected image and measures its similarity with a thumbnail of arbitrary size that is queried by a user. In this way, unscrambling of the image database is not necessary. In this work, the MPEG- 7 Color Layout Descriptor (CLD) is used to represent image features. The matching step is accomplished by calculating the distance between the CLD of the DC image and the CLD of the thumbnail of various sizes. We also investigated the effect of querying thumbnails of various sizes on retrieval performance. We concluded that image size has nearly no impact on the ranks of retrieved images.