“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Sparse coding has become an increasingly popular method in learning and vision for a variety of classification, reconstruction and coding tasks. The canonical approach intrinsically assumes independence between observations during learning. For many natural signals however, sparse coding is applied to sub-elements ( i.e. patches) of the signal, where such an assumption is invalid. Convolutional sparse coding explicitly models local interactions through the convolution operator, however the resulting optimization problem is considerably more complex than traditional sparse coding. In this paper, we draw upon ideas from signal processing and Augmented Lagrange Methods (ALMs) to produce a fast algorithm with globally optimal sub problems and super-linear convergence.

Fast convolutional sparse coding

In this correspondence, we present an accurate Magnetic Resonance (MR) image Super-Resolution (SR) method that uses a Very Deep Residual network (VDR-net) in the training phase. By applying 2D Stationary Wavelet Transform (SWT), we decompose each Low Resolution (LR)-High Resolution (HR) example image pair into its low-frequency and high-frequency subbands. These LR-HR subbands are used to train the VDR-net through the input and output channels. The trained parameters are then used to generate residual subbands of a given LR test image. The obtained residuals are added with their LR subbands to produce the SR subbands. Finally, we attempt to maintain the intrinsic structure of images by implementing the Gaussian edge-preservation step on the SR subbands. Our extensive experimental results show that the proposed MR-SR method outperforms the existing methods in terms of four different objective metrics and subjective quality.

/pdf/accurate-magnetic-resonance-image-super-resolution-using-4l3crjteyv.pdf

Accurate Magnetic Resonance Image Super-Resolution Using Deep Networks and Gaussian Filtering in the Stationary Wavelet Domain

Early detection and diagnosis of coronary artery disease could reduce the risk of developing a heart attack. The coronary arteries are optimally visualised using computed tomography coronary angiography (CTCA) imaging. These images are reviewed by specialist radiologists who evaluate the coronary arteries for potential narrowing. A lack of radiologists in the UK is a constraint to timely diagnosis of coronary artery disease, particularly in the acute accident and emergency department setting. The development of automated methods by which coronary artery narrowing can be identified rapidly and accurately are therefore timely. Such complex computer based tools also need to be sufficiently computationally efficient that they can run on servers typically found in hospital settings, where graphical processing units for example are unavailable. We propose a fully automatic two-dimensional Unet model to segment the aorta and coronary arteries on CTCA images. Two models are trained to segment two regions of interest, (1) the aorta and the coronary arteries or (2) the coronary arteries alone. Our method achieves 91.20% and 88.80% dice similarity coefficient accuracy on regions of interest 1 and 2 respectively. Compared with a semi-automatic segmentation method, our model performs better when segmenting the coronary arteries alone. The performance of the proposed method is comparable to existing published two-dimensional or three-dimensional deep learning models. Importantly, the algorithmic and graphical processing unit memory efficiencies are maintained such that the model can be deployed without requiring graphical processing units, and therefore can be used in a hospital setting.

/pdf/a-computationally-efficient-approach-to-segmentation-of-the-3wgidbtas8.pdf

A Computationally Efficient Approach to Segmentation of the Aorta and Coronary Arteries Using Deep Learning

A Comprehensive Review on Analysis and Implementation of Recent Image Dehazing Methods

ICycleGAN: Single image dehazing based on iterative dehazing model and CycleGAN

This paper proposes a kernel-blending connection approximated by a neural network (KBNN) for image classification. A kernel mapping connection structure, guaranteed by the function approximation theorem, is devised to blend feature extraction and feature classification through neural network learning. First, a feature extractor learns features from the raw images. Next, an automatically constructed kernel mapping connection maps the feature vectors into a feature space. Finally, a linear classifier is used as an output layer of the neural network to provide classification results. Furthermore, a novel loss function involving a cross-entropy loss and a hinge loss is proposed to improve the generalizability of the neural network. Experimental results on three well-known image datasets illustrate that the proposed method has good classification accuracy and generalizability.

/pdf/kernel-blending-connection-approximated-by-a-neural-network-35vicj84hr.pdf

Kernel-blending connection approximated by a neural network for image classification

Automated coronary artery tree segmentation in coronary CTA using a multiobjective clustering and toroidal model-guided tracking method.

Generally, most existing super-resolution (SR) methods do not consider noise, which treats SR reconstruction and denoising as two separate problems and performs separately. However, noise is inevitably introduced in the imaging process. Based on analysis of the degraded model, in this paper, the problems of interpolation and denoising are modeled to estimate the noiseless and missing images under the same framework. By applying local fractal dimension (LFD) into image local feature analysis, a noisy single-image SR method is proposed. For each noisy image, we first construct a rational fractal interpolation model containing scaling factors, which can effectively maintain the inherent properties of the data. Furthermore, the original image structure can be well preserved by applying the interpolation model. Considering the local characteristics of the image, scaling factors are calculated on the basis of the LFDs. Then, through further local feature analysis of the interpolated image, a denoising method based on LFD is proposed for recovering a noiseless image. Finally, a high-quality high-resolution image is obtained. Experimental results demonstrate that our method outperforms the state-of-the-art methods both quantitatively and qualitatively.

/pdf/noisy-single-image-super-resolution-based-on-local-fractal-2aelqoeq0u.pdf

Kai Shao

Papers

ICycleGAN: Single image dehazing based on iterative dehazing model and CycleGAN

Kernel-blending connection approximated by a neural network for image classification

Automated coronary artery tree segmentation in coronary CTA using a multiobjective clustering and toroidal model-guided tracking method.

Noisy Single Image Super-Resolution Based on Local Fractal Feature Analysis