1. Super-Resolution Reconstruction of Transmission Electron Microscopy Images Using Deep Learning Authors:Amit Suveer, Anindya Gupta, Gustaf Kylberg(1), Ida-Maria Sintorn(1)
(1) Vironova AB, Stockholm
In Proceedings: IEEE 16th International Symposium on Biomedical Imaging (ISBI’19), Italy, 2019, pp.
548–551
Abstract: Deep learning techniques have shown promising outcomes in single image super-resolution (SR) reconstruction from noisy and blurry low resolution data. The SR reconstruction can cater the fundamental.
limitations of transmission electron microscopy (TEM) imaging to potentially attain a balance among the trade-offs like imaging-speed, spatial/temporal resolution, and dose/exposure-time, which is often difficult to achieve simultaneously otherwise. In this work, we present a convolutional neural network (CNN) model, utilizing both local and global skip connections, aiming for 4 x SR reconstruction of TEM images. We used exact image pairs of a calibration grid to generate our training and independent testing datasets. The results are compared and discussed using models trained on synthetic (downsampled) and real data from the calibration grid. We also compare the variants of the proposed network with well-known classical interpolations techniques. Finally, we investigate the domain adaptation capacity of the CNN-based model by testing it on TEM images of a cilia sample, having different image characteristics as compared to the calibration-grid.
2. Stochastic Distance Transform
Authors:Johan ¨Ofverstedt, Joakim Lindblad(1), Nataˇsa Sladoje(1)
(1) Mathematical Institute of the Serbian Academy of Sciences and Arts, Belgrade, Serbia
In Proceedings: 21th International Conference on Discrete Geometry for Computer Imagery (DGCI), Lec-ture Notes in Computer Science, Vol. 11414, pp. 75–86, France
DOI: 10.1007/978-3-030-14085-4 7
Abstract: Distance transform (DT) and its many variations are ubiquitous tools for image processing and analysis. In many imaging scenarios, the images of interest are corrupted by noise. This has a strong neg-ative impact on the accuracy of the DT, which is highly sensitive to spurious noise points. In this study, we consider images represented as discrete random sets and observe statistics of DT computed on such representations. We, thus, define a stochastic distance transform (SDT), which has an adjustable robustness to noise. Both a stochastic Monte Carlo method and a deterministic method for computing the SDT are proposed and compared. Through a series of empirical tests, we demonstrate that the SDT is effective not only in improving the accuracy of the computed distances in the presence of noise, but also in improving the performance of template matching and watershed segmentation of partially overlapping objects, which are examples of typical applications where DTs are utilized.
3. On the Effectiveness of Generative Adversarial Networks as HEp-2 Image Augmentation Tool Authors: Tom´aˇs Majtner(1), Buda Baji´c(2), Joakim Lindblad(3), Nataˇsa Sladoje(3), Victoria Blanes-Vidal(1), Esmaeil S. Nadimi(1)
(1) Group of Machine Learning and AI, The Maersk Mc-Kinney Moller Institute, University of Southern Denmark, Odense, Denmark
(2) Faculty of Technical Sciences, University of Novi Sad, Serbia
(3) Mathematical Institute of the Serbian Academy of Sciences and Arts, Belgrade, Serbia
In Proceedings: 21st Scandinavian Conference on Image Analysis (SCIA), LNCS, Vol. 11482, pp. 439–
451, Norrk¨oping, Sweden
DOI: 10.1007/978-3-030-20205-7 36
Abstract: One of the big challenges in the recognition of biomedical samples is the lack of large annotated datasets. Their relatively small size, when compared to datasets like ImageNet, typically leads to problems with efficient training of current machine learning algorithms. However, the recent development of genera-tive adversarial networks (GANs) appears to be a step towards addressing this issue. In this study, we focus on one instance of GANs, which is known as deep convolutio nal generative adversarial network (DCGAN).
It gained a lot of attention recently because of its stability in generating realistic artificial images. Our article explores the possibilities of using DCGANs for generating HEp-2 images. We trained multiple DCGANs and generated several datasets of HEp-2 images. Subsequently, we combined them with traditional aug-mentation and evaluated over three different deep learning configurations. Our article demonstrates high visual quality of generated images, which is also supported by state-of-the-art classification results.
4. Mathematical Morphology on Irregularly Sampled Data Applied to Segmentation of 3D Point Clouds of Urban Scenes
Authors: Teo Asplund, Andr´es Serna(1), Beatriz Marcotegui(2), Robin Strand, Cris L. Luengo Hen-driks(3)
(1) Terra3D, Paris, France
(2) MINES ParisTech, PSL Research University, CMM - Centre for Mathematical Morphology, Fontainebleau, France
(3) Flagship Biosciences Inc., Westminster, USA
In Proceedings: International Symposium on Mathematical Morphology and Its Applications to Signal and Image Processing (ISMM 2019), Lecture Notes in Computer Science, Vol. 11564, pp. 375–387
DOI: 10.1007/978-3-030-20867-7 29
Abstract: This paper proposes an extension of mathematical morphology on irregularly sampled signals to 3D point clouds. The proposed method is applied to the segmentation of urban scenes to show its applica-bility to the analysis of point cloud data. Applying the proposed operators has the desirable side-effect of homogenizing signals that are sampled heterogeneously. In experiments we show that the proposed segmen-tation algorithm yields good results on the Paris-rue-Madame database and is robust in terms of sampling density, i.e. yielding similar labelings for more sparse samplings of the same scene.
5. Face Recognition - A One-Shot Learning Perspective
Authors:Sukalpa Chanda(1), Asish Chakrapani(2),Anders Brun, Anders Hast, Umapada Pal(2), David Doermann(3)
(1) Dept. of Information Technology, Østfold University College, Norway
(2) Computer Vision and Pattern Recognition Unit, Indian Statistical Institute, India (3) Computer Science and Engineering, University at Buffalo, USA
In Proceedings: 15th IEEE Conference on Signal Image Technology and Internet based Systems, pp. 113–
119
Abstract: Ability to learn from a single instance is something unique to the human species and One-shot learning algorithms try to mimic this special capability. On the other hand, despite the fantastic performance of Deep Learning-based methods on various image classification problems, performance often depends hav-ing on a huge number of annotated trainhav-ing samples per class. This fact is certainly a hindrance in deployhav-ing deep neural network-based systems in many real-life applications like face recognition. Furthermore, an addition of a new class to the system will require the need to re-train the whole system from scratch. Nev-ertheless, the prowess of deep learned features could also not be ignored. This research aims to combine the best of deep learned features with a traditional One-Shot learning framework. Results obtained on 2 publicly available datasets are very encouraging achieving over 90% accuracy on 5-way One-Shot tasks, and 84% on 50-way One-Shot problems.
6. Finding Logo and Seal in Historical Document Images - An Object Detection based Approach
Authors:Sukalpa Chanda(1), Prashant Kumar Prasad(2), Anders Hast, Anders Brun, Lasse M˚artensson(3), Umapada Pal(4)
(1) Faculty of Computer Science, Østfold University College, Halden, Norway (2) RCC Institute of Information Technology, Kolkata, India
(3) Dept. of Scandinavian Languages, UU
(4) Computer Vision and Pattern Recognition Unit, Indian Statistical Institute, Kolkata, India In Proceedings: The 5th Asian Conference on Pattern Recognition (ACPR 2019), pp. 821–834 DOI: 10.1007/978-3-030-41404-7 58
Abstract: Logo and Seal serves the purpose of authenticating and referring to the source of a document. This strategy was also prevalent in the medieval period. Different algorithm exists for detection of logo and seal in document images. A close look into the present state-of-the-art methods reveals that those methods were focused toward detection of logo and seal in contemporary document images. However, such methods are likely to underperform while dealing with historical documents. This is due to the fact that historical doc-uments are attributed with additional challenges like extra noise, bleed-through effect, blurred foreground elements and low contrast. The proposed method frames the problem of the logo and seals detection in an object detection framework. Using a deep-learning technique it counters earlier mentioned problems and evades the need for any pre-processing stage like layout analysis and/or binarization in the system pipeline.
The experiments were conducted on historical images from 12th to the 16th century and the results obtained were very encouraging for detecting logo in historical document images. To the best of our knowledge,
this is the first attempt on logo detection in historical document images using an object-detection based approach.
7. Making large collections of handwritten material easily accessible and searchable Authors:Anders Hast, Per Cullhed(1), Ekta Vats, Matteo Abrate(2)
(1) University Library, UU
(2) Institute of Informatics and Telematics, CNR, Pisa, Italy
In Proceedings: Italian Research Conference on Digital Libraries - Digital Libraries: Supporting Open Sci-ence (IRCDL 2019), Communications in Computer and Information SciSci-ence (CCIS), Vol. 988, pp. 18–28 DOI: 10.1007/978-3-030-11226-4 2
Abstract: Libraries and cultural organisations contain a rich amount of digitised historical handwritten ma-terial in the form of scanned images. A vast majority of this mama-terial has not been transcribed yet, owing to technological challenges and lack of expertise. This renders the task of making these historical collections available for public access challenging, especially in performing a simple text search across the collection.
Machine learning based methods for handwritten text recognition are gaining importance these days, which require huge amount of pre-transcribed texts for training the system. However, it is impractical to have ac-cess to several thousands of pre-transcribed documents due to adversities transcribers face. Therefore, this paper presents a training-free word spotting algorithm as an alternative for handwritten text transcription, where case studies on Alvin (Swedish repository) and Clavius on the Web are presented. The main focus of this work is on discussing prospects of making materials in the Alvin platform and Clavius on the Web easily searchable using a word spotting based handwritten text recognition system.
8. Embedded Prototype Subspace Classification: A subspace learning framework Authors:Anders Hast, Mats Lind(1), Ekta Vats
(1) Dept. of Information Technology, UU
In Proceedings: International Conference on Computer Analysis of Images and Patterns CAIP 2019: Com-puter Analysis of Images and Patterns, Lecture Notes in ComCom-puter Science (LNCS), Vol. 11679, pp. 581–
592DOI: 10.1007/978-3-030-29891-3 51
Abstract: Handwritten text recognition is a daunting task, due to complex characteristics of handwritten letters. Deep learning based methods have achieved significant advances in recognizing challenging hand-written texts because of its ability to learn and accurately classify intricate patterns. However, there are some limitations of deep learning, such as lack of well-defined mathematical model, black-box learning mechanism, etc., which pose challenges. This paper aims at going beyond the black-box learning and pro-poses a novel learning framework called as Embedded Prototype Subspace Classification, that is based on the well-known subspace method, to recognise handwritten letters in a fast and efficient manner. The effec-tiveness of the proposed framework is empirically evaluated on popular datasets using standard evaluation measures.
9. Creating an Atlas over Handwritten Script Signs
Authors:Anders Hast, Lasse M˚artensson(1), Ekta Vats, Raphaela Heil (1) Dept. of Swedish Language and Multilingualism, Stockholm University In Proceedings: Digital Humanities in the Nordic Countries
Abstract: A framework for interactive visualization of script characteristics, as present in the form of hand-written letters, is proposed in this work. The basic idea behind this investigation is to lay the foundations for creating a comprehensive atlas over letter forms extracted from a large collection of handwritten documents, with minimal human guidance. The visualization of the results is based on the atlas metaphor and uses the t-SNE visualization method for creating island-like clusters that can be investigated using the proposed vi-sualization framework. By changing a scale parameter one can investigate the dataset on different levels, i.e different sizes of the clusters.
10. Optimization of max-norm objective functions in image processing and computer vision Authors:Filip Malmberg, Krzysztof Ciesielski(1,2),Robin Strand
(1) Dept. of Mathematics, West Virginia University, Morgantown, USA (2) Dept. of Radiology, MIPG University of Pennsylvania, USA
In Proceedings: International Conference on Discrete Geometry for Computer Imagery (DGCI 2019), Lec-ture Notes in Computer Science Vol. 11414, pp. 206–218
DOI: 10.1007/978-3-030-14085-4 17
Abstract: Many fundamental problems in image processing and computer vision, such as image filtering, segmentation, registration, and stereo vision, can naturally be formulated as optimization problems.
We consider binary labeling problems where the objective function is defined as the max-norm over a set of variables. It is well known that for a limited subclass of such problems, globally optimal solutions can be found via watershed cuts, i.e., cuts by optimum spanning forests. Here, we propose a new algorithm for optimizing a broader class of such problems. We prove that the proposed algorithm returns a globally optimal labeling, provided that the objective function satisfies certain given conditions, analogous to the submodularity conditions encountered in min-cut/max-flow optimization. The proposed method is highly efficient, with quasi-linear computational complexity.
11. Distance Transform Based on Weight Sequences
Authors: Benedek Nagy(1),Robin Strand, Nicolas Normand(2)
(1) Eastern Mediterranean University, Famagusta, North Cyprus, via Mersin-10, Turkey (2) Universit´e de Nantes, LS2N UMR CNRS 6004, France
In Proceedings: International Conference on Discrete Geometry for Computer Imagery (DGCI 2019), Lec-ture Notes in Computer Science Vol. 11414, pp. 62–74
DOI: 10.1007/978-3-030-14085-4 6
Abstract: There is a continuous effort to develop the theory and methods for computing digital distance functions, and to lower the rotational dependency of distance functions. Working on the digital space, e.g., on the square grid,digital distance functions are defined by minimal costpaths, which can be processed (back-tracked etc.) without any errors or approximations. Recently, digital distance functions defined by weight sequences, which is a concept allowing multiple types of weighted steps combined with neighbor-hood sequences, were developed. With appropriate weight sequences, the distance between points on the perimeter of a square and the center of the square (i.e., for squares of a given size the weight sequence can be easily computed) are exactly the Euclidean distance for these distances based on weight sequences. How-ever, distances based on weight sequences may not fulfill the triangular inequality. In this paper, continuing the research, we provide a sufficient condition for weight sequences to provide metric distance. Further, we present an algorithm to compute the distance transform based on these distances. Optimization results are also shown for the approximation of the Euclidean distance inside the given square.
12. New Definition of Quality-Scale Robustness for Image Processing Algorithms, with Generalized Un-certainty Modeling, Applied to Denoising and Segmentation
Authors: Antoine Vacavant(1), Marie-Ange Lebre(1), Hugo Rositi(1), Manuel Grand-Brochier(1), Robin Strand
(1) Universit´e Clermont Auvergne, CNRS, SIGMA Clermont, Institut Pascal, Clermont-Ferrand, France In Proceedings: Reproducible Research in Pattern Recognition (RRPR 2018), Lecture Notes in Computer Science, Vol. 11455, pp. 138–149
DOI: 10.1007/978-3-030-23987-9 13
Abstract: Robustness is an important concern in machine learning and pattern recognition, and has attracted a lot of attention from technical and scientific viewpoints. Actually, the robustness models the capacity of a computerized approach to resist to perturbing phenomena and data uncertainties, and generate common artefact while designing algorithms. However, this question has not been dealt in depth in such a way for image processing tasks. In this article, we propose a novel definition of robustness dedicated to image processing algorithms. By considering a generalized model of image data uncertainty, we encompass the classic additive Gaussian noise alteration that we study through the evaluation of image denoising algo-rithms, but also more complex phenomena such as shape variability, which is considered for liver volume segmentation from medical images. Furthermore, we refine our evaluation of robustness wrt. our previous work by introducing a novel quality-scale definition. To do so, we calculate the worst loss of quality for a given algorithm over a set of uncertainty scales, together with the scale where this drop appears. This new approach permits to reveal any algorithm’s weakness, and for which kind of corrupted data it may happen.
13. Segmentation of Post-operative Glioblastoma in MRI by U-Net with Patient-specific Interactive Re-finement
Authors:Ashis Kumar Dhara, Kalyan Ram Ayyalasomayajula, Erik Arvids(1), Markus Fahlstr¨om(1), Johan Wikstr¨om(1), Elna-Marie Larsson(1),Robin Strand
(1) Division of Radiology, Dept. of Surgical Sciences, UU
In Proceedings: 4th International Brain Lesion (BrainLes) workshop, a MICCAI 2018 satellite event. Lec-ture Notes in Computer Science, Vol. 11383 (revised selected papers) pp. 115–122
DOI: 10.1007/978-3-030-11723-8 11
Abstract: Accurate volumetric change estimation of glioblastoma is very important for post-surgical treat-ment follow-up. In this paper, an interactive segtreat-mentation method was developed and evaluated with the aim to guide volumetric estimation of glioblastoma. U-Net based fully convolutional network is used for ini-tial segmentation of glioblastoma from post contrast MR images. The max flow algorithm is applied on the probability map of U-Net to update the initial segmentation and the result is displayed to the user for interac-tive refinement. Network update is performed based on the corrected contour by considering patient specific learning to deal with large context variations among different images. The proposed method is evaluated on a clinical MR image database of 15 glioblastoma patients with longitudinal scan data. The experimental re-sults depict an improvement of segmentation performance due to patient specific fine-tuning. The proposed method is computationally fast and efficient as compared to state-of-the-art interactive segmentation tools.
This tool could be useful for post-surgical treatment follow-up with minimal user intervention.
14. Training-Free and Segmentation-Free Word Spotting using Feature Matching and Query Expansion Authors:Ekta Vats, Anders Hast, Alicia Forn´es(1)
(1) Universitat Aut`onoma de Barcelona
In Proceedings: 15th International Conference on Document Analysis and Recognition, pp. 1294–1299 DOI: 10.1109/ICDAR.2019.00209
Abstract: Historical handwritten text recognition is an interesting yet challenging problem. In recent times, deep learning based methods have achieved significant performance in handwritten text recognition. How-ever, handwriting recognition using deep learning needs training data, and often, text must be previously segmented into lines (or even words). These limitations constrain the application of HTR techniques in doc-ument collections, because training data or segmented words are not always available. Therefore, this paper proposes a training-free and segmentation-free word spotting approach that can be applied in unconstrained scenarios. The proposed word spotting framework is based on document query word expansion and re-laxed feature matching algorithm, which can easily be parallelised. Since handwritten words posses distinct shape and characteristics, this work uses a combination of different keypoint detectors and Fourier-based descriptors to obtain a sufficient degree of relaxed matching. The effectiveness of the proposed method is empirically evaluated on well-known benchmark datasets using standard evaluation measures. The use of informative features along with query expansion significantly contributed in efficient performance of the proposed method.