We propose to combine semantic data and registration algorithms to solve various image processing problems such as denoising, super-resolution and color-correction. It is shown how such new techniques can achieve significant quality enhancement, both visually and quantitatively, in the case of facial image enhancement. Our model assumes prior high quality data of the person to be processed, but no knowledge of the degradation model. We try to overcome the classical processing limits by using semantically-aware patches, with adaptive size and location regions of coherent structure and context, as building blocks. The method is demonstrated on the problem of cellular photography enhancement of dark facial images for different identities, expressions and poses.
Multiview super-resolution image reconstruction (SRIR) is often cast as a resampling problem by merging non-redundant data from multiple low-resolution (LR) images on a finer high-resolution (HR) grid, while inverting the effect of the camera point spread function (PSF). One main problem with multiview methods is that resampling from nonuniform samples (provided by LR images) and the inversion of the PSF are highly nonlinear and ill-posed problems. Non-linearity and ill-posedness are typically overcome by linearization and regularization, often through an iterative optimization process, which essentially trade off the very same information (i.e. high frequency) that we want to recover. We propose a novel point of view for multiview SRIR: Unlike existing multiview methods that reconstruct the entire spectrum of the HR image from the multiple given LR images, we derive explicit expressions that show how the high-frequency spectra of the unknown HR image are related to the spectra of the LR images. Therefore, by taking any of the LR images as the reference to represent the low-frequency spectra of the HR image, one can reconstruct the super-resolution image by focusing only on the reconstruction of the high-frequency spectra. This is very much like single-image methods, which extrapolate the spectrum of one image, except that we rely on information provided by all other views, rather than by prior constraints as in single-image methods (which may not be an accurate source of information). This is made possible by deriving and applying explicit closed-form expressions that define how the local high frequency information that we aim to recover for the reference high resolution image is related to the local low frequency information in the sequence of views. Results and comparisons with recently published state-of-the-art methods show the superiority of the proposed solution.
The Residual Quantization (RQ) framework is revisited where the quantization distortion is being successively reduced in multi-layers. Inspired by the reverse-water-filling paradigm in rate-distortion theory, an efficient regularization on the variances of the codewords is introduced which allows to extend the RQ for very large numbers of layers and also for high dimensional data, without getting over-trained. The proposed Regularized Residual Quantization (RRQ) results in multi-layer dictionaries which are additionally sparse, thanks to the soft-thresholding nature of the regularization when applied to variance-decaying data which can arise from de-correlating transformations applied to correlated data. Furthermore, we also propose a general-purpose pre-processing for natural images which makes them suitable for such quantization. The RRQ framework is first tested on synthetic variance-decaying data to show its efficiency in quantization of high-dimensional data. Next, we use the RRQ in super-resolution of a database of facial images where it is shown that low-resolution facial images from the test set quantized with codebooks trained on high-resolution images from the training set show relevant high-frequency content when reconstructed with those codebooks.
Generative adversarial networks (GANs) have received a tremendous amount of attention in the past few years, and have inspired applications addressing a wide range of problems. Despite its great potential, GANs are difficult to train. Recently, a series of papers (Arjovsky & Bottou, 2017a; Arjovsky et al. 2017b; and Gulrajani et al. 2017) proposed using Wasserstein distance as the training objective and promised easy, stable GAN training across architectures with minimal hyperparameter tuning. In this paper, we compare the performance of Wasserstein distance with other training objectives on a variety of GAN architectures in the context of single image super-resolution. Our results agree that Wasserstein GAN with gradient penalty (WGAN-GP) provides stable and converging GAN training and that Wasserstein distance is an effective metric to gauge training progress.
Magnetic Resonance Imaging (MRI) offers high-resolution \emph{in vivo} imaging and rich functional and anatomical multimodality tissue contrast. In practice, however, there are challenges associated with considerations of scanning costs, patient comfort, and scanning time that constrain how much data can be acquired in clinical or research studies. In this paper, we explore the possibility of generating high-resolution and multimodal images from low-resolution single-modality imagery. We propose the weakly-supervised joint convolutional sparse coding to simultaneously solve the problems of super-resolution (SR) and cross-modality image synthesis. The learning process requires only a few registered multimodal image pairs as the training set. Additionally, the quality of the joint dictionary learning can be improved using a larger set of unpaired images. To combine unpaired data from different image resolutions/modalities, a hetero-domain image alignment term is proposed. Local image neighborhoods are naturally preserved by operating on the whole image domain (as opposed to image patches) and using joint convolutional sparse coding. The paired images are enhanced in the joint learning process with unpaired data and an additional maximum mean discrepancy term, which minimizes the dissimilarity between their feature distributions. Experiments show that the proposed method outperforms state-of-the-art techniques on both SR reconstruction and simultaneous SR and cross-modality synthesis.
In recent years deep learning algorithms have shown extremely high performance on machine learning tasks such as image classification and speech recognition. In support of such applications, various FPGA accelerator architectures have been proposed for convolutional neural networks (CNNs) that enable high performance for classification tasks at lower power than CPU and GPU processors. However, to date, there has been little research on the use of FPGA implementations of deconvolutional neural networks (DCNNs). DCNNs, also known as generative CNNs, encode high-dimensional probability distributions and have been widely used for computer vision applications such as scene completion, scene segmentation, image creation, image denoising, and super-resolution imaging. We propose an FPGA architecture for deconvolutional networks built around an accelerator which effectively handles the complex memory access patterns needed to perform strided deconvolutions, and that supports convolution as well. We also develop a three-step design optimization method that systematically exploits statistical analysis, design space exploration and VLSI optimization. To verify our FPGA deconvolutional accelerator design methodology we train DCNNs offline on two representative datasets using the generative adversarial network method (GAN) run on Tensorflow, and then map these DCNNs to an FPGA DCNN-plus-accelerator implementation to perform generative inference on a Xilinx Zynq-7000 FPGA. Our DCNN implementation achieves a peak performance density of 0.012 GOPs/DSP.
Signal reconstruction from a sample using an orthogonal projector onto a guiding subspace is theoretically well justified, but may be difficult to practically implement. We propose more general guiding operators, which increase signal components in the guiding subspace relative to those in a complementary subspace, e.g., iterative low-pass edge-preserving filters for super-resolution of images. Two examples of super-resolution illustrate our technology: a no-flash RGB photo guided using a high resolution flash RGB photo, and a depth image guided using a high resolution RGB photo.
Most multispectral remote sensors (e.g. QuickBird, IKONOS, and Landsat 7 ETM+) provide low-spatial high-spectral resolution multispectral (MS) or high-spatial low-spectral resolution panchromatic (PAN) images, separately. In order to reconstruct a high-spatial/high-spectral resolution multispectral image volume, either the information in MS and PAN images are fused (i.e. pansharpening) or super-resolution reconstruction (SRR) is used with only MS images captured on different dates. Existing methods do not utilize temporal information of MS and high spatial resolution of PAN images together to improve the resolution. In this paper, we propose a multiframe SRR algorithm using pansharpened MS images, taking advantage of both temporal and spatial information available in multispectral imagery, in order to exceed spatial resolution of given PAN images. We first apply pansharpening to a set of multispectral images and their corresponding PAN images captured on different dates. Then, we use the pansharpened multispectral images as input to the proposed wavelet-based multiframe SRR method to yield full volumetric SRR. The proposed SRR method is obtained by deriving the subband relations between multitemporal MS volumes. We demonstrate the results on Landsat 7 ETM+ images comparing our method to conventional techniques.
Methods based on convolutional neural network (CNN) have demonstrated tremendous improvements on single image super-resolution. However, the previous methods mainly restore images from one single area in the low resolution (LR) input, which limits the flexibility of models to infer various scales of details for high resolution (HR) output. Moreover, most of them train a specific model for each up-scale factor. In this paper, we propose a multi-scale super resolution (MSSR) network. Our network consists of multi-scale paths to make the HR inference, which can learn to synthesize features from different scales. This property helps reconstruct various kinds of regions in HR images. In addition, only one single model is needed for multiple up-scale factors, which is more efficient without loss of restoration quality. Experiments on four public datasets demonstrate that the proposed method achieved state-of-the-art performance with fast speed.
Majority of the face recognition algorithms use query faces captured from uncontrolled, in the wild, environment. Often caused by the cameras limited capabilities, it is common for these captured facial images to be blurred or low resolution. Super resolution algorithms are therefore crucial in improving the resolution of such images especially when the image size is small requiring enlargement. This paper aims to demonstrate the effect of one of the state-of-the-art algorithms in the field of image super resolution. To demonstrate the functionality of the algorithm, various before and after 3D face alignment cases are provided using the images from the Labeled Faces in the Wild (lfw). Resulting images are subject to testing on a closed set face recognition protocol using unsupervised algorithms with high dimension extracted features. The inclusion of super resolution algorithm resulted in significant improved recognition rate over recently reported results obtained from unsupervised algorithms.
Optimal surface segmentation is a state-of-the-art method used for segmentation of multiple globally optimal surfaces in volumetric datasets. The method is widely used in numerous medical image segmentation applications. However, nodes in the graph based optimal surface segmentation method typically encode uniformly distributed orthogonal voxels of the volume. Thus the segmentation cannot attain an accuracy greater than a single unit voxel, i.e. the distance between two adjoining nodes in graph space. Segmentation accuracy higher than a unit voxel is achievable by exploiting partial volume information in the voxels which shall result in non-equidistant spacing between adjoining graph nodes. This paper reports a generalized graph based multiple surface segmentation method with convex priors which can optimally segment the target surfaces in an irregularly sampled space. The proposed method allows non-equidistant spacing between the adjoining graph nodes to achieve subvoxel segmentation accuracy by utilizing the partial volume information in the voxels. The partial volume information in the voxels is exploited by computing a displacement field from the original volume data to identify the subvoxel-accurate centers within each voxel resulting in non-equidistant spacing between the adjoining graph nodes. The smoothness of each surface modeled as a convex constraint governs the connectivity and regularity of the surface. We employ an edge-based graph representation to incorporate the necessary constraints and the globally optimal solution is obtained by computing a minimum s-t cut. The proposed method was validated on 10 intravascular multi-frame ultrasound image datasets for subvoxel segmentation accuracy. In all cases, the approach yielded highly accurate results. Our approach can be readily extended to higher-dimensional segmentations.
Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.
In this work, we investigate the value of uncertainty modeling in 3D super-resolution with convolutional neural networks (CNNs). Deep learning has shown success in a plethora of medical image transformation problems, such as super-resolution (SR) and image synthesis. However, the highly ill-posed nature of such problems results in inevitable ambiguity in the learning of networks. We propose to account for intrinsic uncertainty through a per-patch heteroscedastic noise model and for parameter uncertainty through approximate Bayesian inference in the form of variational dropout. We show that the combined benefits of both lead to the state-of-the-art performance SR of diffusion MR brain images in terms of errors compared to ground truth. We further show that the reduced error scores produce tangible benefits in downstream tractography. In addition, the probabilistic nature of the methods naturally confers a mechanism to quantify uncertainty over the super-resolved output. We demonstrate through experiments on both healthy and pathological brains the potential utility of such an uncertainty measure in the risk assessment of the super-resolved images for subsequent clinical use.
The latest deep learning approaches perform better than the state-of-the-art signal processing approaches in various image restoration tasks. However, if an image contains many patterns and structures, the performance of these CNNs is still inferior. To address this issue, here we propose a novel feature space deep residual learning algorithm that outperforms the existing residual learning. The main idea is originated from the observation that the performance of a learning algorithm can be improved if the input and/or label manifolds can be made topologically simpler by an analytic mapping to a feature space. Our extensive numerical studies using denoising experiments and NTIRE single-image super-resolution (SISR) competition demonstrate that the proposed feature space residual learning outperforms the existing state-of-the-art approaches. Moreover, our algorithm was ranked third in NTIRE competition with 5-10 times faster computational time compared to the top ranked teams. The source code is available on page : https://github.com/iorism/CNN.git
The most sophisticated existing methods to generate 3D isotropic super-resolution (SR) from non-isotropic electron microscopy (EM) are based on learned dictionaries. Unfortunately, none of the existing methods generate practically satisfying results. For 2D natural images, recently developed super-resolution methods that use deep learning have been shown to significantly outperform the previous state of the art. We have adapted one of the most successful architectures (FSRCNN) for 3D super-resolution, and compared its performance to a 3D U-Net architecture that has not been used previously to generate super-resolution. We trained both architectures on artificially downscaled isotropic ground truth from focused ion beam milling scanning EM (FIB-SEM) and tested the performance for various hyperparameter settings. Our results indicate that both architectures can successfully generate 3D isotropic super-resolution from non-isotropic EM, with the U-Net performing consistently better. We propose several promising directions for practical application.
Much of the recent research on solving iterative inference problems focuses on moving away from hand-chosen inference algorithms and towards learned inference. In the latter, the inference process is unrolled in time and interpreted as a recurrent neural network (RNN) which allows for joint learning of model and inference parameters with back-propagation through time. In this framework, the RNN architecture is directly derived from a hand-chosen inference algorithm, effectively limiting its capabilities. We propose a learning framework, called Recurrent Inference Machines (RIM), in which we turn algorithm construction the other way round: Given data and a task, train an RNN to learn an inference algorithm. Because RNNs are Turing complete [1, 2] they are capable to implement any inference algorithm. The framework allows for an abstraction which removes the need for domain knowledge. We demonstrate in several image restoration experiments that this abstraction is effective, allowing us to achieve state-of-the-art performance on image denoising and super-resolution tasks and superior across-task generalization.
Cross-modal image synthesis is a topical problem in medical image computing. Existing methods for image synthesis are either tailored to a specific application, require large scale training sets, or are based on partitioning images into overlapping patches. In this paper, we propose a novel Dual cOnvolutional filTer lEarning (DOTE) approach to overcome the drawbacks of these approaches. We construct a closed loop joint filter learning strategy that generates informative feedback for model self-optimization. Our method can leverage data more efficiently thus reducing the size of the required training set. We extensively evaluate DOTE in two challenging tasks: image super-resolution and cross-modality synthesis. The experimental results demonstrate superior performance of our method over other state-of-the-art methods.
We propose a novel couple mappings method for low resolution face recognition using deep convolutional neural networks (DCNNs). The proposed architecture consists of two branches of DCNNs to map the high and low resolution face images into a common space with nonlinear transformations. The branch corresponding to transformation of high resolution images consists of 14 layers and the other branch which maps the low resolution face images to the common space includes a 5-layer super-resolution network connected to a 14-layer network. The distance between the features of corresponding high and low resolution images are backpropagated to train the networks. Our proposed method is evaluated on FERET data set and compared with state-of-the-art competing methods. Our extensive experimental results show that the proposed method significantly improves the recognition performance especially for very low resolution probe face images (11.4% improvement in recognition accuracy). Furthermore, it can reconstruct a high resolution image from its corresponding low resolution probe image which is comparable with state-of-the-art super-resolution methods in terms of visual quality.
The recent phenomenal interest in convolutional neural networks (CNNs) must have made it inevitable for the super-resolution (SR) community to explore its potential. The response has been immense and in the last three years, since the advent of the pioneering work, there appeared too many works not to warrant a comprehensive survey. This paper surveys the SR literature in the context of deep learning. We focus on the three important aspects of multimedia - namely image, video and multi-dimensions, especially depth maps. In each case, first relevant benchmarks are introduced in the form of datasets and state of the art SR methods, excluding deep learning. Next is a detailed analysis of the individual works, each including a short description of the method and a critique of the results with special reference to the benchmarking done. This is followed by minimum overall benchmarking in the form of comparison on some common dataset, while relying on the results reported in various works.
Volumetric (3d) images are acquired for many scientific and biomedical purposes using imaging methods such as serial section microscopy, CT scans, and MRI. A frequent step in the analysis and reconstruction of such data is the alignment and registration of images that were acquired in succession along a spatial or temporal dimension. For example, in serial section electron microscopy, individual 2d sections are imaged via electron microscopy and then must be aligned to one another in order to produce a coherent 3d volume. State of the art approaches find image correspondences derived from patch matching and invariant feature detectors, and then solve optimization problems that rigidly or elastically deform series of images into an aligned volume. Here we show how fully convolutional neural networks trained with an adversarial loss function can be used for two tasks: (1) synthesis of missing or damaged image data from adjacent sections, and (2) fine-scale alignment of block-face electron microscopy data. Finally, we show how these two capabilities can be combined in order to produce artificial isotropic volumes from anisotropic image volumes using a super-resolution adversarial alignment and interpolation approach.