We analyze behavioral markers of sustained object viewing during free VR museum exploration using 6DoF head motion, gaze dynamics, and pupil diameter (N=20). Viewing episodes were reviewed using retrospective think-aloud (RTA) data, yielding 758 episodes for analysis. In a linear mixed model controlling for subject and exhibit, lower head rotational velocity was associated with longer dwell time (β =-0.280, p<.001). For episodes exceeding 3 seconds, gaze behavior exhibited a temporal shift: dispersion decreased while spatial entropy increased, indicating that gaze converged on a narrower region while becoming more evenly distributed within that region. Event-related pupil responses time-locked to gaze onset were larger for long-dwell than short-dwell episodes (difference=0.107mm, p=.007, d=0.30), but this pupil result should be treated as exploratory. These findings identify behavioral cues associated with sustained viewing that may inform future trigger design in VR museums.

This study investigated the influence of a 3D avatar on viewers’ gaze patterns as they observed historical photographs presented on large-scale 2D displays. We used Head Mount Display as a controlled platform for eye-tracking. We compared gaze data from nine participants across two conditions: a narration-only condition and an avatar-guided condition, where a virtual agent pointed to specific regions of interest on the photograph. Our results show the avatar significantly improved temporal gaze synchrony, reducing the time to first fixation on an ROI by 57% and increasing scanpath similarity among participants. However, the avatar’s effect on the spatial dispersion of attention—measured by 2D entropy and Root Mean Square distance—varied with the photograph’s composition and the avatar’s position, highlighting a trade-off between focused attention and exploratory viewing. These findings suggest that designing effective virtual agents for educational contexts requires balancing the speed of guidance against the desired degree of focused attention.

A recent attention of JPEG format-compliant encryptions has been concentrated to randomize the specific syntaxes of JPEG file interchange format (JFIF) or to flip the signs of quantized discrete cosine transform (QDCT) coefficients. Although the random sign flip (RSF) achieves low-complexity encryption while preserving the JPEG coding efficiency, the texture information represented by nonzero AC coefficients are still easy to reveal for an attacker using replacement attacks. To reinforce the perceptual degradation and the texture protection at the same complexity as the RSF, we propose an AC prediction error propagation-based encryption (AC-WPE) in the JPEG QDCT domain. Since the AC-WPE encrypts the coefficient signs in the residual QDCT domain obtained by a provisional AC predictor such as differential pulse-code modulation (DPCM) and propagates the encrypted prediction errors among the several blocks by the AC reconstructor, the attack resilience is obviously reinforced. Moreover, to suppress the decrease in coding efficiency, we introduce the randomized prediction intervals into the AC predictor/reconstructor. Experiments with JPEG encryption show that the AC-WPE reinforces the perceptual degradation and the resilience to the replacement attacks, while reasonably expensing the bitrate overheads of crypto-compressed images.

We propose a finely tunable JPEG format-compliant perceptual encryption (FE) with two novel strategies: (i) bitcuboid-based encryption (BE) and (ii) exception-free signed binarization (ESB). BE is an intra- and inter-bitplane encryption technique that provides finely tunable perceptual degradation by encrypting constrained subspaces (‘bitcubes’) of a cuboid-shaped bit set (‘bitcuboid’). ESB is a binarization technique that redecimalizes encrypted binary sequences into signed decimal coefficients without any exception-handling by shifting the negative binary sequences one-by-one. BE with ESB (BEESB) is applied to the quantized discrete cosine transform (QDCT) domain in JPEG compression. The results of our first experiment show that the BE attains fine tunability, which means scalability of the perceptual degradation level with a single encryption method, by encrypting bitcubes of various types and sizes combinatorially. The results of our second experiment show that the BEESB suppresses the bitrate overheads and that BEESB with one of the most secure options suppresses approximately 0.80-187.58% more of the bitrate overheads in terms of Bjøntegaard delta (BD)-rate compared with conventional methods except for some ones. The results of our third experiment show that BEESB has high resilience against attacks.

We propose a perceptual encryption within JPEG compression (EWJ). Although some of the conventional EWJs have the `tun-ability,' which is a property of how many perceptual degradation levels can be provided with the single encryption technique, it is insufficient because either strong level of security or low bitrate overhead is regarded as important. The proposed EWJ detects the specific subspaces of the bits in the quantized discrete cosine transform (QDCT) coefficient blocks (`bitcubes'), such that the non-zero bits crowd, and then the bits are permuted within each bitcube, i.e., not only within each bitplane like the conventional EWJs but also between adjacent bitplanes. The experiments show that the bitcube-based EWJ actually provides more flexible tunability than the conventional EWJs, while compromising the relation between the bitrate overhead and the attack robustness.