Bitstream-Based JPEG Encryption in Real-time

Abstract

We propose a framework to encrypt Baseline JPEG files directly at bitstream level, i.e., without the need to recompress them. Our approach enables encrypting more than 25 pictures per second in VGA resolution, allowing real-time operation in typical video surveillance applications. In addition, our approach preserves the length of the bitstream while being completely format-compliant. Furthermore, we show that an attack on the encryption process, which partly relies on AES, is practically infeasible.

Building a post-compression region-of-interest encryption framework for existing video surveillance systems

“…A more detailed description of all encryption steps as well as a thorough security analysis can be found in [1,43]. Note that, although Auer et al [1] describe a full-picture encryption approach, their security analysis is limited to single blocks, so it can be used for our modified approach as well.…”

“…For RoI encryption, we modify the full encryption approach proposed by Auer et al [1]. Thus, in the following sections, we describe the original approach and our modifications, respectively.…”

“…Since face detection is carried out in the image domain, it is necessary to decode the image for this step, which we do using OpenCV. 1 As all other components operate at bit-stream level, i.e. without the necessity to decode image data, the decoding process for face detection can be simplified to processing grey-scale (luminance) data only.…”

“…The encryption approach proposed in [1] processes four 8 × 8-pixel-sized blocks (for typical 4:2:0 YCbCr subsampling [17]) at a time, i.e. it operates on 16 × 16-pixel-sized units.…”

Joint compression and encryption of visual data using orthogonal parametric transforms

“…In the case of spectral methods, we can indicate approaches that are based on, e.g., permutation of DC and AC 1 coefficients within 8 on 8 pixel areas of an image [5,8,9], distribution of DC coefficients within and permutation of AC coefficients between image areas [10], encryption of selected spectral coefficients ordered by an amount of contained visual information [11,12], or divided into bit layers [13,14], and finally approaches that modify spectral coefficients with aid of bitwise operations [15][16][17], linear orthogonal matrices [18], or non-linear mappings [19]. In the group of methods that encrypt data obtained after Huffman entropy coding [20] the typical approaches simply change the order of symbols describing sequences of zeroed AC coefficients after the step of run-length encoding (RLE) [21], or modify the Huffman tables of codes maintaining the lengths of codes what ensures unchanged performance of compression [4]. Some of the mentioned visual data encryption methods found practical applications providing expected results.…”