Probabilistic Intersection-Over-Union for Training and Evaluation of Oriented Object Detectors.

Oriented object detection is a challenging and relatively new problem. Most existing approaches are based on deep learning and explore Oriented Bounding Boxes (OBBs) to represent the objects. They are typically based on adaptations of traditional detectors that work with Horizontal Bounding Boxes (HBBs), which have been exploring IoU-like loss functions to regress the HBBs. However, extending this idea for OBBs is challenging due to complex formulations or requirement for customized backpropagation implementations. Furthermore, using OBBs presents limitations for irregular or roughly circular objects, since the definition of the ideal OBB is an ambiguous and ill-posed problem. In this work, we jointly tackle the problem of training, representing, and evaluating oriented detectors. We explore Gaussian distributions-called Gaussian Bounding Boxes (GBBs)-as fuzzy representations for oriented objects and propose using a similarity metric between two GBBs based on the Hellinger distance. We show that this metric leads to a differentiable closed-form expression that can be directly used as a localization loss term to train OBB object detectors. We also show that GBBs present a natural representation as elliptical regions (called EBBs), which inherently mitigate ambiguity representation for circular objects. Finally, we empirically show that the proposed similarity metric computed between two GBBs strongly correlates with the IoU between the corresponding EBBs, motivating the name Probabilistic Intersection-over-Union (ProbIoU). Our experiments show that results using ProbIoU as a regression loss are competitive with state-of-the-art alternatives without requiring additional hyperparameters or customized implementations, and that ProbIoU is a promising alternative to evaluate oriented object detectors. Our code is available at https://github.com/ProbIOU/.

Omnidirectional 2.5D representation for COVID-19 diagnosis using chest CTs.

The Coronavirus Disease 2019 (COVID-19) has drastically overwhelmed most countries in the last two years, and image-based approaches using computerized tomography (CT) have been used to identify pulmonary infections. Recent methods based on deep learning either require time-consuming per-slice annotations (2D) or are highly data- and hardware-demanding (3D). This work proposes a novel omnidirectional 2.5D representation of volumetric chest CTs that allows exploring efficient 2D deep learning architectures while requiring volume-level annotations only. Our learning approach uses a siamese feature extraction backbone applied to each lung. It combines these features into a classification head that explores a novel combination of Squeeze-and-Excite strategies with Class Activation Maps. We experimented with public and in-house datasets and compared our results with state-of-the-art techniques. Our analyses show that our method provides better or comparable prediction quality and accurately distinguishes COVID-19 infections from other kinds of pneumonia and healthy lungs.

Deep Multi-Scale Feature Learning for Defocus Blur Estimation.

This paper presents an edge-based defocus blur estimation method from a single defocused image. We first distinguish edges that lie at depth discontinuities (called depth edges, for which the blur estimate is ambiguous) from edges that lie at approximately constant depth regions (called pattern edges, for which the blur estimate is well-defined). Then, we estimate the defocus blur amount at pattern edges only, and explore an interpolation scheme based on guided filters that prevents data propagation across the detected depth edges to obtain a dense blur map with well-defined object boundaries. Both tasks (edge classification and blur estimation) are performed by deep convolutional neural networks (CNNs) that share weights to learn meaningful local features from multi-scale patches centered at edge locations. Experiments on naturally defocused images show that the proposed method presents qualitative and quantitative results that outperform state-of-the-art (SOTA) methods, with a good compromise between running time and accuracy.

A Hierarchical Superpixel-Based Approach for DIBR View Synthesis.

View synthesis allows observers to explore static scenes using aligned color images and depth maps captured in a preset camera path. Among the options, depth-image-based rendering (DIBR) approaches have been effective and efficient since only one pair of color and depth map is required, saving storage and bandwidth. The present work proposes a novel DIBR pipeline for view synthesis that properly tackles the different artifacts that arise from 3D warping, such as cracks, disocclusions, ghosts, and out-of-field areas. A key aspect of our contributions relies on the adaptation and usage of a hierarchical image superpixel algorithm that helps to maintain structural characteristics of the scene during image reconstruction. We compare our approach with state-of-the-art methods and show that it attains the best average results in two common assessment metrics under public still-image and video-sequence datasets. Visual results are also provided, illustrating the potential of our technique in real-world applications.

The Subcortical-Allocortical- Neocortical continuum for the Emergence and Morphological Heterogeneity of Pyramidal Neurons in the Human Brain.

Human cortical and subcortical areas integrate emotion, memory, and cognition when interpreting various environmental stimuli for the elaboration of complex, evolved social behaviors. Pyramidal neurons occur in developed phylogenetic areas advancing along with the allocortex to represent 70-85% of the neocortical gray matter. Here, we illustrate and discuss morphological features of heterogeneous spiny pyramidal neurons emerging from specific amygdaloid nuclei, in CA3 and CA1 hippocampal regions, and in neocortical layers II/III and V of the anterolateral temporal lobe in humans. Three-dimensional images of Golgi-impregnated neurons were obtained using an algorithm for the visualization of the cell body, dendritic length, branching pattern, and pleomorphic dendritic spines, which are specialized plastic postsynaptic units for most excitatory inputs. We demonstrate the emergence and development of human pyramidal neurons in the cortical and basomedial (but not the medial, MeA) nuclei of the amygdala with cells showing a triangular cell body shape, basal branched dendrites, and a short apical shaft with proximal ramifications as "pyramidal-like" neurons. Basomedial neurons also have a long and distally ramified apical dendrite not oriented to the pial surface. These neurons are at the beginning of the allocortex and the limbic lobe. "Pyramidal-like" to "classic" pyramidal neurons with laminar organization advance from the CA3 to the CA1 hippocampal regions. These cells have basal and apical dendrites with specific receptive synaptic domains and several spines. Neocortical pyramidal neurons in layers II/III and V display heterogeneous dendritic branching patterns adapted to the space available and the afferent inputs of each brain area. Dendritic spines vary in their distribution, density, shapes, and sizes (classified as stubby/wide, thin, mushroom-like, ramified, transitional forms, "atypical" or complex forms, such as thorny excrescences in the MeA and CA3 hippocampal region). Spines were found isolated or intermingled, with evident particularities (e.g., an extraordinary density in long, deep CA1 pyramidal neurons), and some showing a spinule. We describe spiny pyramidal neurons considerably improving the connectional and processing complexity of the brain circuits. On the other hand, these cells have some vulnerabilities, as found in neurodegenerative Alzheimer's disease and in temporal lobe epilepsy.

Structure and diversity of human dendritic spines evidenced by a new three-dimensional reconstruction procedure for Golgi staining and light microscopy.

BACKGROUND: Different approaches aim to unravel detailed morphological features of neural cells. Dendritic spines are multifunctional units that reflect cellular connectivity, synaptic strength and plasticity. NEW METHOD: A novel three-dimensional (3D) reconstruction procedure is introduced for visualization of dendritic spines from human postmortem brain tissue using brightfield microscopy. The segmentation model was based on thresholding the intensity values of the dendritic spine image along 'z' stacks. We used median filtering and removed false positives. Fine adjustments during image processing confirmed that the reconstructed image of the spines corresponded to the actual original data. RESULTS: Examples are shown for the cortical amygdaloid nucleus and the CA3 hippocampal area. Structure of spine heads and necks was evaluated at different angles. Our 3D reconstruction images display dendritic spines either isolated or in clusters, in a continuum of shapes and sizes, from simple to more elaborated forms, including the presence of spinule and complex 'thorny excrescences'. COMPARISON WITH EXISTING METHODS: The procedure has the advantages already described for the adapted "single-section" Golgi method, since it provides suitable results using human brains fixed in formalin for long time, is relatively easy, requires minimal equipment, and uses an algorithm for 3D reconstruction that provides high quality images and more precise morphological data. CONCLUSION: The procedure described here allows the reliable visualization and study of human dendritic spines with broad applications for normal controls and pathological studies.

Stereo matching and view interpolation based on image domain triangulation.

This paper presents a new approach for stereo matching and view interpolation problems based on triangular tessellations suitable for a linear array of rectified cameras. The domain of the reference image is initially partitioned into triangular regions using edge and scale information, aiming to place vertices along image edges and increase the number of triangles in textured regions. A region-based matching algorithm is then used to find an initial disparity for each triangle, and a refinement stage is applied to change the disparity at the vertices of the triangles, generating a piecewise linear disparity map. A simple post-processing procedure is applied to connect triangles with similar disparities generating a full 3D mesh related to each camera (view), which are used to generate new synthesized views along the linear camera array. With the proposed framework, view interpolation reduces to the trivial task of rendering polygonal meshes, which can be done very fast, particularly when GPUs are employed. Furthermore, the generated views are hole-free, unlike most point-based view interpolation schemes that require some kind of post-processing procedures to fill holes.

A new methodology for detection of counterfeit Viagra® and Cialis® tablets by image processing and statistical analysis.

This paper proposes a new approach for automatic classification of counterfeit Viagra(®) and Cialis(®) tablets using image processing and statistical analysis. A high resolution VSC 5000 is used for image acquisition in a controlled environment, and the combination of a thresholding technique with morphological operators is used to segment the tablet from the background. A statistical model based on the RGB color components of original samples is built, and the detection of counterfeit tablets was performed by checking the adherence of a test sample to the obtained distribution using the Bhattacharyya distance. Our experimental results indicated that counterfeit tablets can be effective detected using the proposed approach.

Adaptive image denoising using scale and space consistency.

This paper proposes a new method for image denoising with edge preservation, based on image multiresolution decomposition by a redundant wavelet transform. In our approach, edges are implicitly located and preserved in the wavelet domain, whilst image noise is filtered out. At each resolution level, the image edges are estimated by gradient magnitudes (obtained from the wavelet coefficients), which are modeled probabilistically, and a shrinkage function is assembled based on the model obtained. Joint use of space and scale consistency is applied for better preservation of edges. The shrinkage functions are combined to preserve edges that appear simultaneously at several resolutions, and geometric constraints are applied to preserve edges that are not isolated. The proposed technique produces a filtered version of the original image, where homogeneous regions appear separated by well-defined edges. Possible applications include image presegmentation, and image denoising.