Model-based neural evaluation and iterative gradient optimization in image restoration and statistical filtering (original) (raw)
Related papers
Weight assignment for adaptive image restoration by neural networks
IEEE Transactions on Neural Networks, 2000
This paper presents a scheme for adaptively training the weights, in terms of varying the regularization parameter, in a neural network for the restoration of digital images. The flexibility of neural-network-based image restoration algorithms easily allow the variation of restoration parameters such as blur statistics and regularization value spatially and temporally within the image. This paper focuses on spatial variation of the regularization parameter. We first show that the previously proposed neural-network method based on gradient descent can only find suboptimal solutions, and then introduce a regional processing approach based on local statistics. A method is presented to vary the regularization parameter spatially. This method is applied to a number of images degraded by various levels of noise, and the results are examined. The method is also applied to an image degraded by spatially variant blur. In all cases, the proposed method provides visually satisfactory results in an efficient way.
Loss Functions for Image Restoration With Neural Networks
IEEE Transactions on Computational Imaging, 2017
Neural networks are becoming central in several areas of computer vision and image processing and different architectures have been proposed to solve specific problems. The impact of the loss layer of neural networks, however, has not received much attention in the context of image processing: the default and virtually only choice is 2. In this paper, we bring attention to alternative choices for image restoration. In particular, we show the importance of perceptually-motivated losses when the resulting image is to be evaluated by a human observer. We compare the performance of several losses, and propose a novel, differentiable error function. We show that the quality of the results improves significantly with better loss functions, even when the network architecture is left unchanged.
Real-Time Image Restoration with an Artificial Neural Network
. We present a neural network that can be applied to image correction in a preprocessing unit. Blur, geometric distortion and unequal brightness distribution are typical for many scanning techniques and can lead to difficulties during further processing of an image. These and other effects of image degradation which frequently appear spacevariant can be considered simultaneously by this approach. In order to calibrate the correcting system the weights of a neural network are trained. Using suitable training patterns and an appropriate optimization criterion for the degraded images, in the result the dimensioned network represents a space-variant filter with a behavior similar to the well-known Wiener filter. The restoration result can be easily altered by the scheme of the learning data generation. Theoretical considerations and examples for 1-D, 2-D and 3-D implementations in soft- and hardware are given. 1. Introduction When operating with image signals, irregularities of the scan...
Incorporating local statistics in image error measurement for adaptive image restoration
Optical Engineering, 2006
This paper presents an image restoration technique incorporating local statistical knowledge in the cost function. Instead of using a conventional grayscale-based error measurement such as the mean squared error, we compare local statistical information about regions in two images using a new error measure. Transient features such as edges and textures are more strongly emphasized than relatively homogeneous regions. With the addition of this local information, we attempt to provide a measure closer to human visual appraisal. We then extend the popular constrained squared-error cost function by incorporating this image error measure. Due to its nonlinear nature, conventional restoration algorithms cannot optimize this cost function efficiently. Therefore we seek an iterative approach. In particular, an extended neural network algorithm is proposed to perform the restoration. It is shown that this technique is efficient, effective, and robust. It compares favorably with other techniques when applied to both grayscale and color images. The results of a subjective survey comparing the proposed algorithm with a more conventional neural network algorithm are presented. The subjects tested in the survey overwhelmingly favored the results provided by the proposed method.
Adaptive image restoration using a local neural approach
2007
This work aims at defining and experimentally evaluating an iterative strategy based on neural learning for blind image restoration in the presence of blur and noise. A salient aspect of our solution is the local estimation of the restored image based on gradient descent strategies able to estimate both the blurring function and the regularized terms adaptively. Instead of explicitly defining the values of local regularization parameters through predefined functions, an adaptive learning approach is proposed. The method was evaluated experimentally using a test pattern generated by a function checkerboard in Matlab. To investigate whether the strategy can be considered an alternative to conventional restoration procedures the results were compared with those obtained by a well known neural restoration approach.
Image Restoration in Neural Network Domain using Back Propagation Network Approach
Image, 2011
Image Restoration is a process by which an image suffering some form of distortion or degradation can be recovered to its original form. Many techniques have been implemented for image restoration for achieving better performance and quality image, often the benefits of improving image quality to the maximum possible extent far outweigh the cost and complexity of the restoration algorithms involved. In this paper, we consider the problem of an image restoration degraded by a blur function and corrupted by noise. Here we are applying the Back Propagation Neural Network approach for image restoration. This method is an iterative approach and attractive because of its improved performance and achieving high quality image in terms of peak signal to noise ratio.
IEEE Transactions on Neural Networks, 2001
In this paper, we address the problem of adaptive regularization in image restoration by adopting a neural-network learning approach. Instead of explicitly specifying the local regularization parameter values, they are regarded as network weights which are then modified through the supply of appropriate training examples. The desired response of the network is in the form of a gray level value estimate of the current pixel using weighted order statistic (WOS) filter. However, instead of replacing the previous value with this estimate, this is used to modify the network weights, or equivalently, the regularization parameters such that the restored gray level value produced by the network is closer to this desired response. In this way, the single WOS estimation scheme can allow appropriate parameter values to emerge under different noise conditions, rather than requiring their explicit selection in each occasion. In addition, we also consider the separate regularization of edges and textures due to their different noise masking capabilities. This in turn requires discriminating between these two feature types. Due to the inability of conventional local variance measures to distinguish these two high variance features, we propose the new edge-texture characterization (ETC) measure which performs this discrimination based on a scalar value only. This is then incorporated into a fuzzified form of the previous neural network which determines the degree of membership of each high variance pixel in two fuzzy sets, the EDGE and TEXTURE fuzzy sets, from the local ETC value, and then evaluates the appropriate regularization parameter by appropriately combining these two membership function values.
Semi-blind image restoration using a local neural approach
Neurocomputing, 2009
This work aims to define and experimentally evaluate an iterative strategy based on neural learning for semi-blind image restoration in the presence of blur and noise. Salient aspects of the proposed strategy are the use of a local error function derived from the conventional global constrained error measure and the assignment of a separate regularization parameter to each image pixel based on local gray level variance. This method can be viewed as a neural strategy where the pixels of the restored image are the synapse's weights the neural network tries to modify during learning to minimize the output error measurement. The method was experimentally evaluated in terms of restoration quality and speed using test images synthetically degraded and increasingly corrupted. To investigate whether the strategy can be considered an alternative to neural restoration procedures, the results were compared with those obtained by well known Hopfield-based restoration approaches. Results obtained show that our method performs significantly better and faster than other models considered.
An artificial neural network for real-time image restoration
Today optical measuring devices are used in many applications. The measurement accuracy should be very good. But when operating with image signals, irregularities of the scanning system must often be corrected. Blur, geometric distortion and unequal brightness distribution can lead to difficulties during further processing of an image. In the following, it is shown how an artificial neural network can be applied to image restoration. In order to calibrate the correcting system the weights of the neural network are trained. Using suitable training patterns and an appropriate optimization criterion for the degraded images, in the result the dimensioned network represents a space variant filter with a behavior similar to the well-known Wiener filter. A pipeline processor simulates a neural network operating in real time. Theoretical considerations and experimental results are given in this this paper
Towards Perceptually Plausible Training of Image Restoration Neural Networks
2019 Ninth International Conference on Image Processing Theory, Tools and Applications (IPTA), 2019
Learning-based black-box approaches have proven to be successful at several tasks in image and video processing domain. Many of these approaches depend on gradient-descent and back-propagation algorithms which requires to calculate the gradient of the loss function. However, many of the visual metrics are not differentiable, and despite their superior accuracy, they cannot be used to train neural networks for imaging tasks. Most of the image restoration neural networks rely on mean squared error to train. In this paper, we investigate visual system based metrics in order to provide perceptual loss functions that can replace mean squared error for gradient descentbased algorithms. We also share our preliminary results on the proposed approach.