In this paper, we present the multi-stage attentive network (MSAN), an efficient and good generalization performance convolutional neural network (CNN) architecture for motion deblurring. We build a multi-stage encoder–decoder network with self-attention and use the binary cross-entropy loss to train our model. In MSAN, there are two core designs. First, we introduce a new attention-based end-to-end method on top of multi-stage networks, which applies group convolution to the self-attention module, effectively reducing the computing cost and improving the model’s adaptability to different blurred images. Secondly, we propose using binary cross-entropy loss instead of pixel loss to optimize our model to minimize the over-smoothing impact of pixel loss while maintaining a good deblurring effect. We conduct extensive experiments on several deblurring datasets to evaluate the performance of our solution for deblurring. Our MSAN achieves superior performance while also generalizing and compares well with state-of-the-art methods.

A novel adaptive secure holographic image watermarking method in the sharp frequency localized contourlet transform (SFLCT) domain is presented. Based upon the sine logistic modulation map and the logistic map, we develop an encrypted binary computer-generated hologram technique to fabricate a hologram of a watermark first. Owing to the enormous key space of the encrypted hologram, the security of the image watermarking system is increased. Then the hologram watermark is embedded into the SFLCT coefficients with Schur decomposition. To obtain better imperceptibility and robustness, the entropy and the edge entropy are utilized to select the suitable watermark embedding positions adaptively. Compared with other watermarking schemes, the suggested method provides a better performance with respect to both imperceptibility and robustness. Experiments show that our watermarking scheme for images is not only is secure and invisible, but also has a stronger robustness against different kinds of attack.

An RGB-D image encryption scheme is presented by combining a newly constructed 4D-ICTM hyperchaotic system with a reformulated quaternion gyrator transform (QGT). The proposed 4D-ICTM dynamics arise from the nonlinear combination of the ICMIC and Chebyshev maps with the tanh function and modulo operation. Their dynamical and statistical properties are examined using Lyapunov exponents, bifurcation diagrams, histograms, autocorrelation analysis, and correlation-dimension computation. To handle the heterogeneous RGB-D data, a dedicated preprocessing procedure converts the RGB channels and 16-bit depth data into four 8-bit components, which are represented jointly in the quaternion domain. The reformulated QGT is evaluated in terms of numerical consistency and computational cost using numerical error, SSIM, and execution-time measurements. The resulting encryption approach is further assessed through statistical analysis, adjacent-pixel correlation, floating-frequency analysis, graphic autocorrelation, and decryption-error evaluation. The experiments show resistance to the tested attack scenarios together with competitive computational performance. An encryption throughput of 43.32 Mbps is obtained at 406 cycles/byte, indicating the potential of the proposed scheme for RGB-D image encryption applications.