Image halftoning and reconstruction using a neural network

Nonlinearity is an important factor in the biological visual neural networks. Among prominent features of the visual networks, movement detections are carried out in the visual cortex. The visual cortex for the movement detection, consist of two layered networks, called the primary visual cortex (V1), followed by the middle temporal area (MT), in which nonlinear functions will play important roles in the visual systems. These networks will be decomposed to asymmetric sub-networks with nonlinearities. In this paper, the fundamental characteristics in asymmetric and symmetric neural networks with nonlinearities are developed for the detection of the changing stimulus or the movement detection in these neural networks. By the optimization of the asymmetric networks, movement detection Equations are derived. Then, it was clarified that the even – odd nonlinearity combined asymmetric networks, has the ability of generating directional vector in the stimulus change detection or movement detection, while symmetric networks need the time memory to have the same ability. Further, the vector operations in the neural network are developed. These facts are applied to two layered networks, V1 and MT.

Digital halftone is the technique used to convert gray-scale images into binary ones, simulating gray shades by scattering appropriately black and white pixels. Sometimes, there arises the necessity of increasing the resolution of a halftone image. Some recent works have proposed a number of learning-based techniques to zoom binary images. However, they cannot consider a large neighborhood to decide the colors of the resolution-increased pixels, because their running time skyrockets with the growth of the window and the sample images sizes. The use of large window and samples is required to zoom halftone images accurately. This paper presents a new technique to zoom quickly and precisely images generated by any locally-decided halftone algorithm. It is based on decision-tree learning and it is very fast, even using a large window or large samples. The zoomed images obtained by this technique are incredibly sharp and accurate.

This paper describes an approach to evaluate the distortions introduced by changing individual pixels in the binary document image. The approach calculates the distortion by combining the change of smoothness with the change of connectivity. It is different from traditional distortion measures, such as PSNR and SNR, which do not correlate with human visual perception for binary image. PSNR and SNR only concern about how many pixels have been changed. The new approach considers not only how many pixels have been changed but also where the pixels have been changed. Simulation results show that the new distortion measure correlates well with human visual perception and its application in watermarking can be effective.

This paper deals not only with pointwise and uniform convergence but also Y-valued fractional approximation results by univariate symmetrized neural network (SNN) operators on Banach space Y,.. Moreover, our main motivation in this work is to compare the convergence results obtained by classical neural network (NN) operators and symmetric neural network (SNN) operators and try to convert them into numerical examples and graphs through computer programming language Python codes. As a result of this experimental study conducted under the regime of certain parameters, the convergence speed and results of SNN operators are superior to those of classical NN operators.

The author discusses a problem of motion perception in the neural network of the catfish retina. Linear and nonlinear information are crucial to visual processing in the catfish retinal network. Horizontal and bipolar cell responses are linearly related to the input modulation of light, while amacrine cells work as a nonlinear function of squaring. These cells make an asymmetrical nonlinear network. The author solves a problem of motion perception in the catfish retinal-like network. It is shown that this asymmetrical neural network is sensitive to the directional moving stimulus. The conditions of the directional movement are shown by the equations of Wiener kernels in the asymmetrical neural network. It is also shown that a symmetrical neural network cannot determine the equations for the moving stimulus. >

The classical interchange (permutation) of atoms of similar identity does not have an effect on the overall potential energy. In this study, we present feed-forward neural network structures that provide permutation symmetry to the potential energy surfaces of molecules. The new feed-forward neural network structures are employed to fit the potential energy surfaces for two illustrative molecules, which are H(2)O and ClOOCl. Modifications are made to describe the symmetric interchange (permutation) of atoms of similar identity (or mathematically, the permutation of symmetric input parameters). The combined-function-derivative approximation algorithm (J. Chem. Phys. 2009, 130, 134101) is also implemented to fit the neural-network potential energy surfaces accurately. The combination of our symmetric neural networks and the function-derivative fitting effectively produces PES fits using fewer numbers of training data points. For H(2)O, only 282 configurations are employed as the training set; the testing root-mean-squared and mean-absolute energy errors are respectively reported as 0.0103 eV (0.236 kcal/mol) and 0.0078 eV (0.179 kcal/mol). In the ClOOCl case, 1693 configurations are required to construct the training set; the root-mean-squared and mean-absolute energy errors for the ClOOCl testing set are 0.0409 eV (0.943 kcal/mol) and 0.0269 eV (0.620 kcal/mol), respectively. Overall, we find good agreements between ab initio and NN prediction in term of energy and gradient errors, and conclude that the new feed-forward neural-network models advantageously describe the molecules with excellent accuracy.

Visual motion information is useful for many complex tasks in the biological and robotic systems. Models for motion processing in the biological systems have been studied to use conventional symmetric quadrature functions with Gabor filters. This paper proposes a model of the another bio-inspired asymmetric neural networks. The prominent features are the nonlinear characteristics as the squaring and rectification functions, which are observed in the retinal and visual cortex networks. In this paper, the asymmetric network with Gabor filters is compared with that of the conventional symmetric networks. It is shown that the biological asymmetric network with nonlinearities is effective for detecting the inputted phase information and directional movements from the network computations. The responses to the frequency characteristics and to the complex motion stimulus are computed in the asymmetric networks, which are not derived for the conventional energy model.

Face recognition is done gray or color images. Face recognition can be also done in binary images. Halftone images are binary images having gray shade information. These images are more informative as compared to binary images. Halftone images can be generated from different kernels. The quality of a halftone image depends on the kernel used to generate it. Halftone images can be used to generate human recognizable features for face recognition. The human recognizable features are fed to multiclass SVM classifiers for face recognition. The performance of using different human recognizable features generated from halftone images of different standard kernels is studied. It is found that some of kernels give higher recognition rate than the other. Also, the performance is also dependent on size of the window used for feature extraction. In most cases, the minimum size window 3 × 3 gives higher recognition rate in most of the cases. Of the standard kernels Frankie Siera-3, Stucki kernels give higher recognition rate and of the proposed kernels P2 kernel gives higher recognition rate than other proposed kernels.

A hit-or-miss transform architecture and a canonical Boolean logic format and threshold decomposition algorithm are proposed to process binary and gray-tone images with morphological combined operations. The transformation of morphological combined image processing functions to hit-or-miss transforms is explicit: Binary functions are formulated by a summation of hit-or-miss transforms and gray-tone functions formulated by a double summation. A simple optical hit-or-miss transform processor is suggested on the basis of incoherent two-channel correlation with dual-rail spatial coding. The hit-or-miss transforms can be realized optically in one parallel processing step. Thus, binary and gray-tone image processing functions can be implemented by the repeated use of the optical processor with different structuring elements.

The authors study temporal association in a stochastic neural network model with symmetric full-connections. A symmetric system is accessible to analysis because of the existence of free-energy. The properties of the model are analytically described by critical temperature of transition between states. The result of the analysis is consistent with Monte Carlo simulations. >

In 1984 Hopfield showed that the time evolution of a symmetric Hopfield neural networks are a motion in state space that seeks out minima in the energy function (i.e., equilibrium point set of Hopfield neural networks). Because high-order Hopfield neural networks have more extensive applications than Hopfield neural networks, and have been discussed on the convergence of the networks. In practice, a neural network is often subject to environmental noise. It is therefore useful and interesting to find out whether the high-order neural network system still approacher some limit set under stochastic perturbation. In this paper, we will give a number of useful bounds for the noise intensity under which the stochastic high-order neural network will approach its limit set. Our result cancels the requirement of symmetry of the connection weight matrix and includes the classic result on Hopfield neural networks, which is a special case of stochastic high-order Hopfield neural networks. In the end, A example is given to verify the effective of our results.

Data hiding and reversible data hiding research has primarily focused on grayscale and color images, because binary and halftone images are prone to visual distortion caused by a small number of errors in pixel representation. As a result, reversible data hiding is more useful than halftone-based data hiding. This study proposes an investigation of encrypted halftone images based on dual reversible data hiding, which improves the reversibility and security of the image by utilizing a dual cover image. Since halftone images are adequately compressed, they are beneficial in low-channel-bandwidth environments. Hamming code (HC) (7,4) is applied to each block of the halftone image to hide the secret data, and two halftone images are recorded and sent to different receivers at the end of the embedding process. Recipients can use the proposed method and the two marked images to extract the message and recover the cover halftone image. The proposed data hiding method can enhance the quality of the decrypted image by appropriately increasing the block size, and conversely, sufficiently large amounts of data can be hidden by reducing the block size. The experimental results provide evidence of the effectiveness of the proposed method in terms of both image quality and the embedding rate.

Introduction. Working with graphic images is an essential element of almost any modern computer-aided design systems. The result of neural network, including deep image processing, can be recognition of the belonging classes of objects present on them. Objects of the real world require large expenditures for the development and implementation of highly specialized computer vision systems. In recent years, there has been an improvement of the quality characteristics obtained in the field of technical vision. This is made possible by artificial neural networks. The article deals with the recognition of a flat, black and white flowchart image. These are two-dimensional images or their selected parts, which are displayed in an arbitrary graphic format by system means on a computer monitor screen. Basic block diagram shapes: rectangle, rhombus, parallelogram, circle, ellipse (oval), etc. The purpose of the article is to solve the problem of graphic image recognition. The systems for the recognition of graphic images include modern systems of computer-aided design, management and document management. The article has formed a basic set of training and test images of block diagram nodes. Neural network models are proposed to improve the detection accuracy of block diagram nodes based on fully connected and convolutional neural networks. Results. The basic procedures of the block diagram image recognition algorithm were partially tested at the software level, and allow us to conclude the effectiveness of the proposed structural methods. A comparative analysis of neural network and syntactic structural approaches for solving this problem is carried out. Conclusions. Two methods of recognizing flat graphic figures and recognizing connections between figures in flowcharts are proposed: a method of recursive traversal of all branches of the tree of the current union of connections between figures and the figures themselves, and also a study was carried out for the created neural network in PyTorch to solve this problem using trained neural network methods. Keywords: image recognition, convolutional neural networks, syntactic analysis.

The objective of this work is to reconstruct high quality gray-level images from halftone images, or the inverse halftoning process. We develop high performance halftone reconstruction methods for several commonly used halftone techniques. For better reconstruction quality, image classification based on halftone techniques is placed before the reconstruction process so that the halftone reconstruction process can be fine tuned for each halftone technique. The classification is based on enhanced 1D correlation of halftone images and processed with a three- layer back propagation neural network. This classification method reached 100 percent accuracy with a limited set of images processed by dispersed-dot ordered dithering, clustered-dot ordered dithering, constrained average, and error diffusion methods in our experiments. For image reconstruction, we apply the least-mean-square adaptive filtering algorithm which intends to discover the optimal filter weights and the mask shapes. As a result, it yields very good reconstruction image quality. The error diffusion yields the best reconstructed quality among the halftone methods. In addition, the LMS method generates optimal image masks which are significantly different for each halftone method. These optimal masks can also be applied to more sophisticated reconstruction methods as the default filter masks.

We study the q -state Potts-glass neural network with the pseudoinverse (PI) rule. Its performance is investigated and compared with that of the counterpart network with the Hebbian rule instead. We find that there exists a critical point of q , i.e., q(cr)=14 , below which the storage capacity and the retrieval quality can be greatly improved by introducing the PI rule. We show that the dynamics of the neural networks constructed with the two learning rules respectively are quite different; but however, regardless of the learning rules, in the q-state Potts-glass neural networks with q≥3 there is a common novel dynamical phase in which the spurious memories are completely suppressed. This property has never been noticed in the symmetric feedback neural networks. Free from the spurious memories implies that the multistate Potts-glass neural networks would not be trapped in the metastable states, which is a favorable property for their applications.