A light field (LF) image is composed of multiple sub-aperture images (SAIs) from slightly offset viewpoints. To denoise a noise-corrupted LF image, leveraging recent development in deep algorithm unfolding, we pursue a hybrid graph-model-based / data-driven approach. Specifically, we first connect each pixel in a target patch of an SAI to neighboring pixels within the patch, and to pixels in co-located "similar" patches in adjacent SAIs. Given graph connectivity, we formulate a maximum a posteriori (MAP) problem using graph total variation (GTV) as signal prior. We then unroll the iterations of a corresponding optimization algorithm into a sequence of neural layers. In each unrolled layer, we learn relevant features per pixel from data using a convolutional neural net (CNN) in a supervised manner, so that edge weights can be computed as functions of feature distances. Each neural layer can be interpreted as a graph low-pass filter for a 4D LF image patch. Experiments show that our proposal outperformed two model-based and two deep-learning-based implementations in numerical and visual comparisons.

While deep learning (DL) architectures like convolutional neural networks (CNNs) have enabled effective solutions in image denoising, in general their implementations overly rely on training data, lack interpretability, and require tuning of a large parameter set. In this paper, we combine classical graph signal filtering with deep feature learning into a competitive hybrid design—one that utilizes interpretable analytical low-pass graph filters and employs 80% fewer network parameters than state-of-the-art DL denoising scheme DnCNN. Specifically, to construct a suitable similarity graph for graph spectral filtering, we first adopt a CNN to learn feature representations per pixel, and then compute feature distances to establish edge weights. Given a constructed graph, we next formulate a convex optimization problem for denoising using a graph total variation (GTV) prior. Via a l1graph Laplacian reformulation, we interpret its solution in an iterative procedure as a graph low-pass filter and derive its frequency response. For fast filter implementation, we realize this response using a Lanczos approximation. Experimental results show that in the case of statistical mistmatch, our algorithm outperformed DnCNN by up to 3dB in PSNR.

Adithya V Ganesan, Matthew Matero, Aravind Reddy Ravula, Huy Vu, H. Andrew Schwartz. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

While relational algebra and calculus are a well-established foundation for classical database query languages, it is less clear what the analog is for higher-order functions, such as query transformations. Here we study a natural way to add higher-order functionality to query languages, by adding database query operators to the λ-calculus as constants. This framework, which we refer to as λ-embedded query languages, was introduced in [BPV10]. That work had a restricted focus: the containment and equivalence problems for query-to-query functions, in the case where only positive relational operators are allowed as constants. In this work we take an in-depth look at the most basic issue for such languages: the evaluation problem. We give a full picture of the complexity of evaluation for λ-embedded query languages, looking at a number of variations: with negation and without; with only relational algebra operators, and also with a recursion mechanism in the form of a query iteration operator; in a strongly-typed calculus as well as a weakly-typed one. We give tight bounds on both the combined complexity and the query complexity of evaluation in all these settings, in the process explaining connections with Datalog and prior work on λ-calculus evaluation.

We investigate a higher-order query language that embeds operators of the positive relational algebra within the simply-typed λ-calculus. Our language allows one to succinctly define ordinary positive relational algebra queries (conjunctive queries and unions of conjunctive queries) and, in addition, second-order query functionals, which allow the transformation of CQs and UCQs in a generic (i.e., syntax-independent) way. We investigate the equivalence and containment problems for this calculus, which subsumes traditional CQ/UCQ containment. Query functionals are said to be equivalent if the output queries are equivalent, for each possible input query, and similarly for containment. These notions of containment and equivalence depend on the class of (ordinary relational algebra) queries considered. We show that containment and equivalence are decidable when query variables are restricted to positive relational algebra and we identify the precise complexity of the problem. We also identify classes of functionals where containment is tractable. Finally, we provide upper bounds to the complexity of the containment problem when functionals act over other classes.