This paper proposes a new neural network structure for image processing whose convolutional layers, instead of using kernels with fixed coefficients, use space-variant coefficients. The adoption of this strategy allows the system to adapt its behavior according to the spatial characteristics of the input data. This type of layers performs, as we demonstrate, a non-linear transfer function. The features generated by these layers, compared to the ones generated by canonical CNN layers, are more complex and more suitable to fit to the local characteristics of the images. Networks composed by these non-linear layers offer performance comparable with or superior to the ones which use canonical Convolutional Networks, using fewer layers and a significantly lower number of features. Several applications of these newly conceived networks to classical image-processing problems are analyzed. In particular, we consider: Single-Image Super-Resolution (SISR), Edge-Preserving Smoothing (EPS), Noise Removal (NR), and JPEG artifacts removal (JAR).

This paper presents an image enhancement model, D2BGAN (Dark to Bright Generative Adversarial Network), to translate low light images to bright images without a paired supervision. We introduce the use of geometric and lighting consistency along with a contextual loss criterion. These when combined with multiscale color, texture and edge discriminators prove to provide competitive results. We performed extensive experiments using benchmark datasets to visually and objectively compare our results. We observed the performance of D2BGAN on real-time driving datasets that are subject to motion blur, noise, and other artifacts. We further demonstrated that our enhanced images can be profitably used in image-understanding tasks. Images processed using our technique obtain the best or second best average scores for three different image quality evaluation methods on the Naturalness Preserved Enhancement (NPE), Low Light Image Enhancement (LIME), Multi-Exposure Image Fusion (MEF) benchmark datasets. Best scores are also obtained on the LOw-Light (LOL) test set and on Berkeley Driving Dataset (BDD) images processed with D2BGAN. Face detection tasks on the DarkFace benchmark dataset show an mAP (mean Average Precision) improvement from 0.209 to 0.301 when images are processed using D2BGAN. mAP further improves to 0.525 when finetuning techniques are adopted.

In the recent years, there is an upsurge of Artificial Intelligent (AI) systems. These systems, along with efficient performance and predictability also need to incorporate the power of explainability and interpretability. This can significantly aid clinical decision support by providing explainable predictions to assist clinicians. Explainability generally involves uncovering of key input features important for classification. However, characterizing the uncertainty underlying the decisions of the AI system is an important aspect needed for interpreting the decisions. This is especially important in clinical decision support systems given considerations of medical ethics such as nonmaleficence and beneficence. In this study, we develop methods for characterizing the decision certainty of Machine Learning (ML) based clinical decision support systems. As an illustrative example, we introduce a framework for ML based posttraumatic stress disorder (PTSD) diagnostic classification which classifies the subjects into pure and mixed classes. Accordingly, a clinician can have very high confidence (≥ 95% probability) about the diagnosis of a subject in a pure PTSD or combat control class. Remaining sample points for which the AI classification tool does not have very high confidence (< 95% probability) are grouped into a mixed class. Such a scheme will address ethical considerations of nonmaleficence and beneficence since the clinicians can use the AI system to identify those subjects whose diagnosis has very high degree of confidence (and proceed treatment accordingly), and refer those in the uncertain/mixed group to further tests. This is a novel approach, in contrast to existing framework which aim to maximize classification.