Separable 2D Lifting Using Discrete-Time Cellular Neural Networks for Lossless Image Coding

Journal of Signal Processing

et al. 2014

Cycle spinning cellular neural networks (CS-CNNs) are artificial neural networks that work effectively to solve large-scale problems. In our previous work, a CS-CNN is applied to enhance the resolution of images with an arbitrary magnification parameter. In this paper, a novel adaptive architecture using a CS-CNN is developed to prevent the unnecessary smoothing of image detail. While a discrete-time cellular neural network (DT-CNN) transforms all pixel values into coefficients to predict the original pixel values using the A-template, the adaptive cycle spinning method with "minmod" functions is applied to estimate the optimal coefficients from individual outputs of the DT-CNN as above. The minmod functions are defined on the basis of the interpolation error theorem. Experimental results indicate that the proposed method produces better results than the conventional image resolution enhancement methods.

Section: Cycle Spinning Techniquementioning

confidence: 97%

Section: Introductionmentioning

confidence: 99%

Adaptive Cycle Spinning Cellular Neural Network for Image Resolution Enhancement

Chiba

Journal of Signal Processing

et al. 2014

“…The discrete-time cellular neural networks (DT-CNNs) [1], [2]has been applied to many applications such as image compression, filtering and recognition [3]- [10]. Especially, the image processing tasks are its best application for utilizing the nonlinear spatio-temporal dynamics.…”

Section: Introductionmentioning

confidence: 99%

“…Hence, it is possible to build wavelets by neuro dynamic algorithm. Moreover, in [10], the DT-CNNs is successfully applied for lifting scheme for lossless image coding. However, since the stability point of nonlinear Lyapunov energy function is depends to the initial state value of dynamics cells, the intermittent chaos lead to an annealing effect is applied to proposed method.…”

Section: Introductionmentioning

confidence: 99%

An annealing method for cellular neural networks

Konishi

2010 12th International Workshop on Cellular Nanoscale Networks and Their Applications (CNNA 2010)

et al. 2010

The spurious minima in optimizing operation is one of the difficulty for Lyapunov function. In this paper, novel lossless image coding method based on lifting scheme using discrete-time cellular neural networks (DT-CNNs) with annealing approach is proposed. In the proposed, the image prediction of lifting scheme is implimented by DT-CNNs solving the nonlinear optimization problem of Lyapunov energy function. Since the stability point of DT-CNNs energy function is depends to the initial state value of cells, an annealing effect of adaptive chaotic noise is used to avoid the difficulty of global asymptotical stability of DT-CNNs dynamics. The experimental results show that the proposed method produces better results than those of conventional lossless image coding methods.

“…Because of this characteristic, its best application is image processing tasks such as image compression, filtering, learning, pattern recognition [4]- [7]. As discussed in many literatures [3], [8]- [11], CNN designing method and stability conditions have become clearness, and the alternatives of its output function, which determines the dynamic range of CNN stable equilibrium output, has become more flexible; candidates of the output function are piece wise linear (PWL) functions for onebit output, and a quantization function for any level output.…”

Section: Introductionmentioning

confidence: 99%

Hardware-aware model of sigma-delta cellular neural network

Yuki

2009 European Conference on Circuit Theory and Design

et al. 2009

The sigma-delta cellular neural network (SD-CNN) is a complete framework of a spatial domain sigma-delta modulator, and has a very high image reconstruction (AD-to-DA) performance. In this architecture, the A-template, given by a 2D low pass filter (LPF), is used for a digital to analogue converter (DAC), the C-template works as an integrator, and the nonlinear output function is for the bilevel output. By exploiting to the nonlinear optimization ability of CNN spatiotemporal dynamics, optimal binary and reconstruction image can be obtained. However, in the conventional SD-CNN, the Gaussian LPF, whose coefficients are real number, is used as the A-template. This filter coefficients requirement is one of major factors that restricts a hardware implementation. In this paper, a SD-CNN having hardware-friendly filter coefficients is proposed. Moreover its AD and DA performance is confirmed by some experiments.