Probabilistic Deterministic Finite Automata and Recurrent Networks, Revisited

Reservoir computers (RCs) and recurrent neural networks (RNNs) can mimic any finite-state automaton in theory, and some workers demonstrated that this can hold in practice. We test the capability of generalized linear models, RCs, and Long Short-Term Memory (LSTM) RNN architectures to predict the stochastic processes generated by a large suite of probabilistic deterministic finite-state automata (PDFA) in the small-data limit according to two metrics: predictive accuracy and distance to a predictive rate-distortion curve. The latter provides a sense of whether or not the RNN is a lossy predictive feature extractor in the information-theoretic sense. PDFAs provide an excellent performance benchmark in that they can be systematically enumerated, the randomness and correlation structure of their generated processes are exactly known, and their optimal memory-limited predictors are easily computed. With less data than is needed to make a good prediction, LSTMs surprisingly lose at predictive accuracy, but win at lossy predictive feature extraction. These results highlight the utility of causal states in understanding the capabilities of RNNs to predict.

A Complexity Reduction Method for VVC Intra Prediction Based on Statistical Analysis and SAE-CNN

Compared with High Efficiency Video Coding (HEVC), the latest video coding standard Versatile Video Coding Standard (VVC), due to the introduction of many novel technologies and the introduction of the Quad-tree with nested Multi-type Tree (QTMT) division scheme in the block division method, the coding quality has been greatly improved. Due to the introduction of the QTMT scheme, the encoder needs to perform rate–distortion optimization for each division mode during Coding Unit (CU) division, so as to select the best division mode, which also leads to an increase in coding time and coding complexity. Therefore, we propose a VVC intra prediction complexity reduction algorithm based on statistical theory and the Size-adaptive Convolutional Neural Network (SAE-CNN). The algorithm combines the establishment of a pre-decision dictionary based on statistical theory and a Convolutional Neural Network (CNN) model based on adaptively adjusting the size of the pooling layer to form an adaptive CU size division decision process. The algorithm can make a decision on whether to divide CUs of different sizes, thereby avoiding unnecessary Rate–distortion Optimization (RDO) and reducing coding time. Experimental results show that compared with the original algorithm, our suggested algorithm can save 35.60% of the coding time and only increases the Bjøntegaard Delta Bit Rate (BD-BR) by 0.91%.

Improved Coefficient Recovery and Its Application for Rewritable Data Embedding

JPEG is the most commonly utilized image coding standard for storage and transmission purposes. It achieves a good rate–distortion trade-off, and it has been adopted by many, if not all, handheld devices. However, often information loss occurs due to transmission error or damage to the storage device. To address this problem, various coefficient recovery methods have been proposed in the past, including a divide-and-conquer approach to speed up the recovery process. However, the segmentation technique considered in the existing method operates with the assumption of a bi-modal distribution for the pixel values, but most images do not satisfy this condition. Therefore, in this work, an adaptive method was employed to perform more accurate segmentation, so that the real potential of the previous coefficient recovery methods can be unleashed. In addition, an improved rewritable adaptive data embedding method is also proposed that exploits the recoverability of coefficients. Discrete cosine transformation (DCT) patches and blocks for data hiding are judiciously selected based on the predetermined precision to control the embedding capacity and image distortion. Our results suggest that the adaptive coefficient recovery method is able to improve on the conventional method up to 27% in terms of CPU time, and it also achieved better image quality with most considered images. Furthermore, the proposed rewritable data embedding method is able to embed 20,146 bits into an image of dimensions 512×512.

Time Delay Optimization of Compressing Shipborne Radar Digital Video Based on Deep Learning

The High Efficiency Video Coding Standard (HEVC) is one of the most advanced coding schemes at present, and its excellent coding performance is highly suitable for application in the navigation field with limited bandwidth. In recent years, the development of emerging technologies such as screen sharing and remote control has promoted the process of realizing the virtual driving of unmanned ships. In order to improve the transmission and coding efficiency during screen sharing, HEVC proposes a new extension scheme for screen content coding (HEVC-SCC), which is based on the original coding framework. SCC has improved the performance of compressing computer graphics content and video by adding new coding tools, but the complexity of the algorithm has also increased. At present, there is no delay in the compression optimization method designed for radar digital video in the field of navigation. Therefore, our paper starts from the perspective of increasing the speed of encoded radar video, and takes reducing the computational complexity of the rate distortion cost (RD-cost) as the goal of optimization. By analyzing the characteristics of shipborne radar digital video, a fast encoding algorithm for shipborne radar digital video based on deep learning is proposed. Firstly, a coding tree unit (CTU) division depth interval dataset of shipborne radar images was established. Secondly, in order to avoid erroneously skipping of the intra block copy (IBC)/palette mode (PLT) in the coding unit (CU) division search process, we designed a method to divide the depth interval by predicting the CTU in advance and limiting the CU rate distortion cost to be outside the traversal calculation depth interval, which effectively reduced the compression time. The effect of radar transmission and display shows that, within the acceptable range of Bjøntegaard Delta Bit Rate (BD-BR) and Bjøntegaard Delta Peak Signal to Noise Rate (BD-PSNR) attenuation, the algorithm proposed in this paper reduces the coding time by about 39.84%, on average, compared to SCM8.7.

Compression and Visualization Interactive of 3D Mesh

The combination of compression and visualization is mentioned as perspective, very few articles treat with this problem. Indeed, in this paper, we proposed a new approach to multiresolution visualization based on a combination of segmentation and multiresolution mesh compression. For this, we proposed a new segmentation method that benefits the organization of faces of the mesh followed by a progressive local compression of regions of mesh to ensure the refinement local of the three-dimensional object. Thus, the quantization precision is adapted to each vertex during the encoding /decoding process to optimize the rate-distortion compromise. The optimization of the treated mesh geometry improves the approximation quality and the compression ratio at each level of resolution. The experimental results show that the proposed algorithm gives competitive results compared to the previous works dealing with the rate-distortion compromise and very satisfactory visual results.

Learned Hyperspectral Compression Using a Student’s T Hyperprior

Hyperspectral compression is one of the most common techniques in hyperspectral image processing. Most recent learned image compression methods have exhibited excellent rate-distortion performance for natural images, but they have not been fully explored for hyperspectral compression tasks. In this paper, we propose a trainable network architecture for hyperspectral compression tasks, which not only considers the anisotropic characteristic of hyperspectral images but also embeds an accurate entropy model using the non-Gaussian prior knowledge of hyperspectral images and nonlinear transform. Specifically, we first design a spatial-spectral block, involving a spatial net and a spectral net as the base components of the core autoencoder, which is more consistent with the anisotropic hyperspectral cubes than the existing compression methods based on deep learning. Then, we design a Student’s T hyperprior that merges the statistics of the latents and the side information concepts into a unified neural network to provide an accurate entropy model used for entropy coding. This not only remarkably enhances the flexibility of the entropy model by adjusting various values of the degree of freedom, but also leads to a superior rate-distortion performance. The results illustrate that the proposed compression scheme supersedes the Gaussian hyperprior universally for virtually all learned natural image codecs and the optimal linear transform coding methods for hyperspectral compression. Specifically, the proposed method provides a 1.51% to 59.95% average increase in peak signal-to-noise ratio, a 0.17% to 18.17% average increase in the structural similarity index metric and a 6.15% to 64.60% average reduction in spectral angle mapping over three public hyperspectral datasets compared to the Gaussian hyperprior and the optimal linear transform coding methods.

Mathematical Modeling for Ceramic Shape 3D Image Based on Deep Learning Algorithm

Ceramic image shape 3D image modeling focuses on of ceramic that was obtained from the camera imaging equipment such as 2D images, by normalization, gray, filtering denoising, wavelet image sharpening edge enhancement, binarization, and shape contour extraction pretreatment processes such as extraction ceramic image shape edge profile, again, according to the image edge extraction and elliptic rotator ceramics phenomenon. The image distortion effect was optimized by self-application, and then the deep learning modeler was used to model the side edge contour. Finally, the 3D ceramic model of the rotating body was restored according to the intersection and central axis of the extracted contour. By studying the existing segmentation methods based on deep learning, the automatic segmentation of target ceramic image and the effect of target edge refinement and optimization are realized. After extracting and separating the target ceramics from the image, we processed the foreground image of the target into a three-dimensional model. In order to reduce the complexity of the model, a 3D contextual sequencing model is adopted to encode the hidden space features along the channel dimensions, to extract the causal correlation between channels. Each module in the compression framework is optimized by a rate-distortion loss function. The experimental results show that the proposed 3D image modeling method has significant advantages in compression performance compared with the optimal 2D 3D image modeling method based on deep learning, and the experimental results show that the performance of the proposed method is superior to JP3D and HEVC methods, especially at low bit rate points.