Semi-supervised atmospheric component learning in low-light image problem.

Ambient lighting conditions play a crucial role in determining the perceptual quality of images from photographic devices. In general, inadequate transmission light and undesired atmospheric conditions jointly degrade the image quality. If we know the desired ambient factors associated with the given low-light image, we can recover the enhanced image easily. Typical deep networks perform enhancement mappings without investigating the light distribution and color formulation properties. This leads to a lack of image instance-adaptive performance in practice. On the other hand, physical model-driven schemes suffer from the need for inherent decompositions and multiple objective minimizations. Moreover, the above approaches are rarely data efficient or free of postprediction tuning. Influenced by the above issues, this study presents a semisupervised training method using no-reference image quality metrics for low-light image restoration. We incorporate the classical haze distribution model to explore the physical properties of the given image to learn the effect of atmospheric components and minimize a single objective for restoration. We validate the performance of our network for six widely used low-light datasets. Experimental studies show that our proposed study achieves a competitive performance for no-reference metrics compared to current state-of-the-art methods. We also show the improved generalization performance of our proposed method which is efficient in preserving face identities in extreme low-light scenarios.

Enhancing Ensemble Learning Using Explainable CNN for Spoof Fingerprints.

Convolutional Neural Networks (CNNs) have demonstrated remarkable success with great accuracy in classification problems. However, the lack of interpretability of the predictions made by neural networks has raised concerns about the reliability and robustness of CNN-based systems that use a limited amount of training data. In such cases, the utilization of ensemble learning using multiple CNNs has demonstrated the capability to improve the robustness of a network, but the robustness can often have a trade-off with accuracy. In this paper, we propose a novel training method that utilizes a Class Activation Map (CAM) to identify the fingerprint regions that influenced previously trained networks to attain their predictions. The identified regions are concealed during the training of networks with the same architectures, thus enabling the new networks to achieve the same objective from different regions. The resultant networks are then ensembled to ensure that the majority of the fingerprint features are taken into account during classification, resulting in significant enhancement of classification accuracy and robustness across multiple sensors in a consistent and reliable manner. The proposed method is evaluated on LivDet datasets and is able to achieve state-of-the-art accuracy.

Can hippocampal subfield measures supply information that could be used to improve the diagnosis of Alzheimer's disease?

The diagnosis of Alzheimer's disease (AD) needs to be improved. We investigated if hippocampal subfield volume measured by structural imaging, could supply information, so that the diagnosis of AD could be improved. In this study, subjects were classified based on clinical, neuropsychological, and amyloid positivity or negativity using PET scans. Data from 478 elderly Korean subjects grouped as cognitively unimpaired ß-amyloid-negative (NC), cognitively unimpaired ß-amyloid-positive (aAD), mild cognitively impaired ß-amyloid-positive (pAD), mild cognitively impaired-specific variations not due to dementia ß-amyloid-negative (CIND), severe cognitive impairment ß-amyloid-positive (ADD+) and severe cognitive impairment ß-amyloid-negative (ADD-) were used. NC and aAD groups did not show significant volume differences in any subfields. The CIND did not show significant volume differences when compared with either the NC or the aAD (except L-HATA). However, pAD showed significant volume differences in Sub, PrS, ML, Tail, GCMLDG, CA1, CA4, HATA, and CA3 when compared with the NC and aAD. The pAD group also showed significant differences in the hippocampal tail, CA1, CA4, molecular layer, granule cells/molecular layer/dentate gyrus, and CA3 when compared with the CIND group. The ADD- group had significantly larger volumes than the ADD+ group in the bilateral tail, SUB, PrS, and left ML. The results suggest that early amyloid depositions in cognitive normal stages are not accompanied by significant bilateral subfield volume atrophy. There might be intense and accelerated subfield volume atrophy in the later stages associated with the cognitive impairment in the pAD stage, which subsequently could drive the progression to AD dementia. Early subfield volume atrophy associated with the ß-amyloid burden may be characterized by more symmetrical atrophy in CA regions than in other subfields. We conclude that the hippocampal subfield volumetric differences from structural imaging show promise for improving the diagnosis of Alzheimer's disease.

Rethinking Gradient Weight's Influence over Saliency Map Estimation.

Class activation map (CAM) helps to formulate saliency maps that aid in interpreting the deep neural network's prediction. Gradient-based methods are generally faster than other branches of vision interpretability and independent of human guidance. The performance of CAM-like studies depends on the governing model's layer response and the influences of the gradients. Typical gradient-oriented CAM studies rely on weighted aggregation for saliency map estimation by projecting the gradient maps into single-weight values, which may lead to an over-generalized saliency map. To address this issue, we use a global guidance map to rectify the weighted aggregation operation during saliency estimation, where resultant interpretations are comparatively cleaner and instance-specific. We obtain the global guidance map by performing elementwise multiplication between the feature maps and their corresponding gradient maps. To validate our study, we compare the proposed study with nine different saliency visualizers. In addition, we use seven commonly used evaluation metrics for quantitative comparison. The proposed scheme achieves significant improvement over the test images from the ImageNet, MS-COCO 14, and PASCAL VOC 2012 datasets.

Denoising Single Images by Feature Ensemble Revisited.

Image denoising is still a challenging issue in many computer vision subdomains. Recent studies have shown that significant improvements are possible in a supervised setting. However, a few challenges, such as spatial fidelity and cartoon-like smoothing, remain unresolved or decisively overlooked. Our study proposes a simple yet efficient architecture for the denoising problem that addresses the aforementioned issues. The proposed architecture revisits the concept of modular concatenation instead of long and deeper cascaded connections, to recover a cleaner approximation of the given image. We find that different modules can capture versatile representations, and a concatenated representation creates a richer subspace for low-level image restoration. The proposed architecture's number of parameters remains smaller than in most of the previous networks and still achieves significant improvements over the current state-of-the-art networks.

Ensemble of ROI-based convolutional neural network classifiers for staging the Alzheimer disease spectrum from magnetic resonance imaging.

Patches from three orthogonal views of selected cerebral regions can be utilized to learn convolutional neural network (CNN) models for staging the Alzheimer disease (AD) spectrum including preclinical AD, mild cognitive impairment due to AD, and dementia due to AD and normal controls. Hippocampi, amygdalae and insulae were selected from the volumetric analysis of structured magnetic resonance images (MRIs). Three-view patches (TVPs) from these regions were fed to the CNN for training. MRIs were classified with the SoftMax-normalized scores of individual model predictions on TVPs. The significance of each region of interest (ROI) for staging the AD spectrum was evaluated and reported. The results of the ensemble classifier are compared with state-of-the-art methods using the same evaluation metrics. Patch-based ROI ensembles provide comparable diagnostic performance for AD staging. In this work, TVP-based ROI analysis using a CNN provides informative landmarks in cerebral MRIs and may have significance in clinical studies and computer-aided diagnosis system design.

Fast Single-Image HDR Tone-Mapping by Avoiding Base Layer Extraction.

The tone-mapping algorithm compresses the high dynamic range (HDR) information into the standard dynamic range for regular devices. An ideal tone-mapping algorithm reproduces the HDR image without losing any vital information. The usual tone-mapping algorithms mostly deal with detail layer enhancement and gradient-domain manipulation with the help of a smoothing operator. However, these approaches often have to face challenges with over enhancement, halo effects, and over-saturation effects. To address these challenges, we propose a two-step solution to perform a tone-mapping operation using contrast enhancement. Our method improves the performance of the camera response model by utilizing the improved adaptive parameter selection and weight matrix extraction. Experiments show that our method performs reasonably well for overexposed and underexposed HDR images without producing any ringing or halo effects.

Automatic aortic valve landmark localization in coronary CT angiography using colonial walk.

The minimally invasive transcatheter aortic valve implantation (TAVI) is the most prevalent method to treat aortic valve stenosis. For pre-operative surgical planning, contrast-enhanced coronary CT angiography (CCTA) is used as the imaging technique to acquire 3-D measurements of the valve. Accurate localization of the eight aortic valve landmarks in CT images plays a vital role in the TAVI workflow because a small error risks blocking the coronary circulation. In order to examine the valve and mark the landmarks, physicians prefer a view parallel to the hinge plane, instead of using the conventional axial, coronal or sagittal view. However, customizing the view is a difficult and time-consuming task because of unclear aorta pose and different artifacts of CCTA. Therefore, automatic localization of landmarks can serve as a useful guide to the physicians customizing the viewpoint. In this paper, we present an automatic method to localize the aortic valve landmarks using colonial walk, a regression tree-based machine-learning algorithm. For efficient learning from the training set, we propose a two-phase optimized search space learning model in which a representative point inside the valvular area is first learned from the whole CT volume. All eight landmarks are then learned from a smaller area around that point. Experiment with preprocedural CCTA images of TAVI undergoing patients showed that our method is robust under high stenotic variation and notably efficient, as it requires only 12 milliseconds to localize all eight landmarks, as tested on a 3.60 GHz single-core CPU.

Geodesic Distance Algorithm for Extracting the Ascending Aorta from 3D CT Images.

This paper presents a method for the automatic 3D segmentation of the ascending aorta from coronary computed tomography angiography (CCTA). The segmentation is performed in three steps. First, the initial seed points are selected by minimizing a newly proposed energy function across the Hough circles. Second, the ascending aorta is segmented by geodesic distance transformation. Third, the seed points are effectively transferred through the next axial slice by a novel transfer function. Experiments are performed using a database composed of 10 patients' CCTA images. For the experiment, the ground truths are annotated manually on the axial image slices by a medical expert. A comparative evaluation with state-of-the-art commercial aorta segmentation algorithms shows that our approach is computationally more efficient and accurate under the DSC (Dice Similarity Coefficient) measurements.

A Novel Cascade Classifier for Automatic Microcalcification Detection.

In this paper, we present a novel cascaded classification framework for automatic detection of individual and clusters of microcalcifications (µC). Our framework comprises three classification stages: i) a random forest (RF) classifier for simple features capturing the second order local structure of individual µCs, where non-µC pixels in the target mammogram are efficiently eliminated; ii) a more complex discriminative restricted Boltzmann machine (DRBM) classifier for µC candidates determined in the RF stage, which automatically learns the detailed morphology of µC appearances for improved discriminative power; and iii) a detector to detect clusters of µCs from the individual µC detection results, using two different criteria. From the two-stage RF-DRBM classifier, we are able to distinguish µCs using explicitly computed features, as well as learn implicit features that are able to further discriminate between confusing cases. Experimental evaluation is conducted on the original Mammographic Image Analysis Society (MIAS) and mini-MIAS databases, as well as our own Seoul National University Bundang Hospital digital mammographic database. It is shown that the proposed method outperforms comparable methods in terms of receiver operating characteristic (ROC) and precision-recall curves for detection of individual µCs and free-response receiver operating characteristic (FROC) curve for detection of clustered µCs.

Forest Walk Methods for Localizing Body Joints from Single Depth Image.

We present multiple random forest methods for human pose estimation from single depth images that can operate in very high frame rate. We introduce four algorithms: random forest walk, greedy forest walk, random forest jumps, and greedy forest jumps. The proposed approaches can accurately infer the 3D positions of body joints without additional information such as temporal prior. A regression forest is trained to estimate the probability distribution to the direction or offset toward the particular joint, relative to the adjacent position. During pose estimation, the new position is chosen from a set of representative directions or offsets. The distribution for next position is found from traversing the regression tree from new position. The continual position sampling through 3D space will eventually produce an expectation of sample positions, which we estimate as the joint position. The experiments show that the accuracy is higher than current state-of-the-art pose estimation methods with additional advantage in computation time.