This article presents a configurable, high-throughput pseudo-random number generator template targeting cryptographic applications. The template is parameterized using a chaotic map that generates data, an entropy builder that is used to periodically change the parameters of the map and a parameter change interval, which is the number of iterations after which the entropy builder will change the generator’s parameters. The system is implemented in C++ and evaluated using the TestU01 and NIST RNG statistical tests. The same implementation is used for a stream cipher that can encrypt and decrypt PNG images. A Monte-Carlo analysis of the seed space was performed. Results show that for certain combinations of maps and entropy builders, more than 90% of initial states (seeds) tested pass all statistical randomness tests. Also, the throughput is large enough so that a 8 K color image can be encrypted in 2 s on a modern laptop CPU (exact specifications are given in the paper). The conclusion is that chaotic maps can be successfully used as a building block for cryptographic random number generators.
This paper presents an improved image encryption algorithm using Haar wavelet packets decomposition and four chaotic maps. For the wavelet packets scrambling we propose a permutation based on the Baker chaotic map. To increase the entropy of the ciphered image and the key space of the cryptosystem two chaotic maps – the three-dimensional Arnold map, applied twice, and the tent map combined with the logistic map – are used. Matlab simulations show that, in each step, these chaotic maps increase the entropy of the encrypted images and the pixels’ decorrelation. Another benefit obtained using the four maps is the augmentation of the key space.
The novelty of this work is the statistical perspective of the cryptanalyst being able to perform a chosen plaintext attack on a private communication scheme involving chaos-based algorithms. The attacker finds himself in the position of analyzing the pseudo-random matrix that was used for encryption. The contribution is done in the context of the well-known fields of secret-key cryptology and that of wavelet packet decomposition of images. The analysis is exemplified on the thoroughly investigated, in the existing literature, simplest chaotic system, the logistic map. Kolmogorov-Smirnov tests, histogram testing, autocorrelation functions and nonlinear singular value decomposition are used. In addition, a new enciphering wavelet-based algorithm is proposed and analyzed.
Emotion recognition in video (ERV) aims to infer human affect from visual, audio, and contextual signals and is increasingly important for interactive and intelligent systems. Over the past decade, ERV has evolved from handcrafted features and task-specific deep learning models toward transformer-based vision–language models and multimodal large language models (MLLMs). This review surveys this evolution, with an emphasis on engineering considerations relevant to real-world deployment. We analyze multimodal fusion strategies, dataset characteristics, and evaluation protocols, highlighting limitations in robustness, bias, and annotation quality under unconstrained conditions. Emerging MLLM-based approaches are examined in terms of performance, reasoning capability, computational cost, and interaction potential. By comparing task-specific models with foundation model approaches, we clarify their respective strengths for resource-constrained versus context-aware applications. Finally, we outline practical research directions toward building robust, efficient, and deployable ERV systems for applied scenarios such as assistive technologies and human–AI interaction.