Self-organized operational neural networks for severe image restoration problems
نویسندگان
چکیده
Discriminative learning based on convolutional neural networks (CNNs) aims to perform image restoration by from training examples of noisy-clean pairs. It has become the go-to methodology for tackling and outperformed traditional non-local class methods. However, top-performing are generally composed many layers hundreds neurons, with trainable parameters in excess several million. We claim that this is due inherently linear nature convolution-based transformation, which inadequate handling severe problems. Recently, a non-linear generalization CNNs, called operational (ONN), been shown outperform CNN AWGN denoising. its formulation burdened fixed collection well-known operators an exhaustive search find best possible configuration given architecture, whose efficacy further limited output layer operator assignment. In study, we leverage Taylor series-based function approximation propose self-organizing variant ONNs, Self-ONNs, restoration, synthesizes novel nodal transformations on-the-fly as part process, thus eliminating need redundant runs search. addition, it enables finer level heterogeneity diversifying individual connections receptive fields weights. series extensive ablation experiments across three tasks. Even when strict equivalence learnable imposed, Self-ONNs surpass CNNs considerable margin all problems, improving performance up 3 dB terms PSNR.
منابع مشابه
Self-organized critical neural networks.
A mechanism for self-organization of the degree of connectivity in model neural networks is studied. Network connectivity is regulated locally on the basis of an order parameter of the global dynamics, which is estimated from an observable at the single synapse level. This principle is studied in a two-dimensional neural network with randomly wired asymmetric weights. In this class of networks,...
متن کاملBayesian Neural Networks for Image Restoration
Numerical methods commonly employed to convert experimental data into interpretable images and spectra commonly rely on straightforward transforms, such as the Fourier transform (FT), or quite elaborated emerging classes of transforms, like wavelets (Meyer, 1993; Mallat, 2000), wedgelets (Donoho, 1996), ridgelets (Candes, 1998), and so forth. Yet experimental data are incomplete and noisy due t...
متن کاملEfficient image restoration using cellular neural networks
In this paper, a 3-D Cellular Neural Network (CNN) is applied for restoration of degraded images. It is known that regularized or Maximum a Posteriori estimation based image restoration problems can be formulated as the minimization of the Lyapunov function of the discrete-time Hopeld network. Recently, this Lyapunov function based design method has been extended to the continuous-time Hopeld n...
متن کاملSelf-Commmittee Approach for Image Restoration Problems using Convolutional Neural Network
There have been many discriminative learning methods using convolutional neural networks (CNN) for several image restoration problems, which learn the mapping function from a degraded input to the clean output. In this letter, we propose a self-committee method that can find enhanced restoration results from the multiple trial of a trained CNN with different but related inputs. Specifically, it...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Neural Networks
سال: 2021
ISSN: ['1879-2782', '0893-6080']
DOI: https://doi.org/10.1016/j.neunet.2020.12.014