[go: up one dir, main page]
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
4.1
1.8
Journals
Algorithms
Special Issues
Bio-Inspired Algorithms for Image Processing
share announcement

Bio-Inspired Algorithms for Image Processing

A special issue of Algorithms (ISSN 1999-4893). This special issue belongs to the section "Evolutionary Algorithms and Machine Learning".

Deadline for manuscript submissions: closed (31 October 2020) | Viewed by 29198

Special Issue Editors

Dr. Sándor Szénási

E-Mail Website
Guest Editor
Software Engineering Institute, John von Neumann Faculty of Informatics, Óbuda University, 1034 Budapest, Hungary
Interests: machine learning, neural networks, simulation, GPU programming
Special Issues, Collections and Topics in MDPI journals
Dr. Gábor Kertész

E-Mail
Guest Editor
Software Engineering Institute, John von Neumann Faculty of Informatics, Óbuda University, 1034 Budapest, Hungary
Interests: machine learning; deep neural networks; parallel programming
Special Issues, Collections and Topics in MDPI journals

Special Issue Information

Dear Colleagues,

In the field of image processing, there are several hard problems without exact solutions due to incomplete or imperfect information and limited computation capacity. Some methods of the different subfields (classification, feature extraction, pattern recognition, segmentation, etc.) are based on the process of natural selection, the behavior of living creatures, or especially on the mechanisms of the brain. It is also worth mentioning that in the case of traditional procedural image processing methods, where the algorithms are well-defined, the parametrization can be challenging, if not impossible. This step can be formalized as an optimization problem, where the application of heuristics is necessary for addressing such highly complex problems to provide feasible solutions in acceptable runtime. Many of these optimization techniques are also inspired by nature.

In this Special Issue of "Bio-inspired Algorithms for Image Processing", we seek original research or results of practical applications from the area of bio-inspired algorithms in the field of image processing. We are waiting for manuscripts discussing evolutional (genetic algorithms, NSGA, etc.), swarm intelligence-based (particle swarm optimization, ant colony optimization, fireworks, etc.) and brain-inspired computing (neural networks, deep learning, etc.) methods applied in any kind of image processing research projects (classification, segmentation, medical image processing, etc.).

Dr. Sandor Szenasi
Dr. Gábor Kertész
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Algorithms is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 1600 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

Keywords

  • Design and analysis of bio-inspired methods
  • Application of bio-inspired methods
  • Limitations of bio-inspired methods
  • Ant colony optimization
  • Particle swarm optimization
  • Firefly algorithm
  • Fireworks algorithm
  • Bees algorithm
  • Evolutionary algorithms
  • Neural networks
  • Deep learning
  • Soft computing methods
  • Nature-inspired heuristics

Benefits of Publishing in a Special Issue

  • Ease of navigation: Grouping papers by topic helps scholars navigate broad scope journals more efficiently.
  • Greater discoverability: Special Issues support the reach and impact of scientific research. Articles in Special Issues are more discoverable and cited more frequently.
  • Expansion of research network: Special Issues facilitate connections among authors, fostering scientific collaborations.
  • External promotion: Articles in Special Issues are often promoted through the journal's social media, increasing their visibility.
  • e-Book format: Special Issues with more than 10 articles can be published as dedicated e-books, ensuring wide and rapid dissemination.

Further information on MDPI's Special Issue polices can be found here.

Published Papers (7 papers)

Download All Papers
Order results
Result details
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:

Editorial

Jump to: Research

2 pages, 631 KiB  
Editorial
Special Issue on Bio-Inspired Algorithms for Image Processing
by Sándor Szénási and Gábor Kertész
Algorithms 2020, 13(12), 320; https://doi.org/10.3390/a13120320 - 3 Dec 2020
Viewed by 1859
Abstract
In the field of image processing, there are several difficult issues that do not have exact solutions due to incomplete or imperfect information and limited computation capacity [...] Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)

Research

Jump to: Editorial

27 pages, 14281 KiB  
Article
Pavement Defect Segmentation in Orthoframes with a Pipeline of Three Convolutional Neural Networks
by Roland Lõuk, Andri Riid, René Pihlak and Aleksei Tepljakov
Algorithms 2020, 13(8), 198; https://doi.org/10.3390/a13080198 - 14 Aug 2020
Cited by 11 | Viewed by 3913
Abstract
In the manuscript, the issue of detecting and segmenting out pavement defects on highway roads is addressed. Specifically, computer vision (CV) methods are developed and applied to the problem based on deep learning of convolutional neural networks (ConvNets). A novel neural network structure [...] Read more.
In the manuscript, the issue of detecting and segmenting out pavement defects on highway roads is addressed. Specifically, computer vision (CV) methods are developed and applied to the problem based on deep learning of convolutional neural networks (ConvNets). A novel neural network structure is considered, based on a pipeline of three ConvNets and endowed with the capacity for context awareness, which improves grid-based search for defects on orthoframes by considering the surrounding image content—an approach, which essentially draws inspiration from how humans tend to solve the task of image segmentation. Also, methods for assessing the quality of segmentation are discussed. The contribution also describes the complete procedure of working with pavement defects in an industrial setting, involving the workcycle of defect annotation, ConvNet training and validation. The results of ConvNet evaluation provided in the paper hint at a successful implementation of the proposed technique. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Figure 1

Figure 1
<p>The orthoframes depicting the road.</p> Full article ">Figure 2
<p>Generation of the region-of-interest orthoframe mask.</p> Full article ">Figure 3
<p>Overview of the inference process for defect segmentation over an orthoframe.</p> Full article ">Figure 4
<p>Illustration of Road Segmentation Network (RSN) architecture.</p> Full article ">Figure 5
<p>Road area image pre-processing.</p> Full article ">Figure 6
<p>Extraction of a content-context sample from an orthoframe.</p> Full article ">Figure 7
<p>Architecture of the defect detection convolutional neural network (ConvNet) used in this work.</p> Full article ">Figure 8
<p>Architecture of the defect segmentation U-Net used in this work.</p> Full article ">Figure 9
<p>An illustration of various metrics. (<b>a</b>) The original image; (<b>b</b>) TP, FN and FP pixels are depicted by colors of green, blue and red, respectively; (<b>c</b>) Three objects obtained by combining the predictions and ground truth instances; (<b>d</b>) TP, FN and FP pixels with edge-tolerant intersection over union (IoU) computation.</p> Full article ">Figure 10
<p>Classification of segmented defects. As in <a href="#algorithms-13-00198-f009" class="html-fig">Figure 9</a>, colors green, blue and red depict TP, FN and FP pixels, respectively. (<b>a</b>) true positive; (<b>b</b>) false positive; (<b>c</b>) false negative; (<b>d</b>) false negative and false positive; (<b>e</b>) false positive; (<b>f</b>) false negative;</p> Full article ">Figure 11
<p>Simplified UML diagram of the class and package relationships making up the DATM annotation tool.</p> Full article ">Figure 12
<p>Graphical user interface of the annotation tool written in Python 3 with the PyQt framework.</p> Full article ">Figure 13
<p>Active learning of road contours and pavement defects.</p> Full article ">Figure 14
<p>Three of the sampling strategies visualized on an orthoframe. Each green square is of size 336 × 336 and added to the given dataset. For illustrative purposes, annotated defects are marked in blue and predicted false positives (only shown in c) are marked in red. For the sake of visual clarity, partial overlapping is not displayed and the focused part of the road in conjunction with the road area mask is highlighted. (<b>a</b>) Full sampling.; (<b>b</b>) Dilated defect contour sampling.; (<b>c</b>) Defect contour sampling with false positives.</p> Full article ">Figure 15
<p>Defect Segmentation Network (DSN) results for different encoder architectures over 40 epochs trained on the 200 orthoframe dataset. Performed with 2-fold cross-validation.</p> Full article ">Figure 16
<p>Visualization of the context captured for different DSN patch sizes. For models with input sizes of 336 × 336 or 448 × 448, the output gets cropped to the central part of size 224 × 224. In each case, the orthoframe still gets partitioned into patches of size 224 × 224, which are shown as yellow squares.</p> Full article ">Figure 17
<p>Orthoframes from the test set processed by the PDDS . RSN outputs multiplied by the region of interest mask are highlighted. Patches classified by the DDN as non-defective have white borders and patches classified as defective have red borders. For the masks produced by DSN, the green, red and blue contours represent true positive, false positive and false negative pixels respectively. (<b>a</b>) IoU = 0.846, eIoU = 0.930, iIoU = 0.457, cPr = 0.600, cRc = 0.750; (<b>b</b>) IoU = 0.790, eIoU = 0.991, iIoU = 0.582, cPr = 1, cRc = 1; (<b>c</b>) IoU = 0.104, eIoU = 0.110, iIoU = 0.306, cPr = 0.143, cRc = 1; (<b>d</b>) IoU = 0.515, eIoU = 0.575, iIoU = 0.306, cPr = 0.750, cRc = 0.500.</p> Full article ">
22 pages, 5363 KiB  
Article
Biologically Inspired Visual System Architecture for Object Recognition in Autonomous Systems
by Dan Malowany and Hugo Guterman
Algorithms 2020, 13(7), 167; https://doi.org/10.3390/a13070167 - 11 Jul 2020
Cited by 4 | Viewed by 4536
Abstract
Computer vision is currently one of the most exciting and rapidly evolving fields of science, which affects numerous industries. Research and development breakthroughs, mainly in the field of convolutional neural networks (CNNs), opened the way to unprecedented sensitivity and precision in object detection [...] Read more.
Computer vision is currently one of the most exciting and rapidly evolving fields of science, which affects numerous industries. Research and development breakthroughs, mainly in the field of convolutional neural networks (CNNs), opened the way to unprecedented sensitivity and precision in object detection and recognition tasks. Nevertheless, the findings in recent years on the sensitivity of neural networks to additive noise, light conditions, and to the wholeness of the training dataset, indicate that this technology still lacks the robustness needed for the autonomous robotic industry. In an attempt to bring computer vision algorithms closer to the capabilities of a human operator, the mechanisms of the human visual system was analyzed in this work. Recent studies show that the mechanisms behind the recognition process in the human brain include continuous generation of predictions based on prior knowledge of the world. These predictions enable rapid generation of contextual hypotheses that bias the outcome of the recognition process. This mechanism is especially advantageous in situations of uncertainty, when visual input is ambiguous. In addition, the human visual system continuously updates its knowledge about the world based on the gaps between its prediction and the visual feedback. CNNs are feed forward in nature and lack such top-down contextual attenuation mechanisms. As a result, although they process massive amounts of visual information during their operation, the information is not transformed into knowledge that can be used to generate contextual predictions and improve their performance. In this work, an architecture was designed that aims to integrate the concepts behind the top-down prediction and learning processes of the human visual system with the state-of-the-art bottom-up object recognition models, e.g., deep CNNs. The work focuses on two mechanisms of the human visual system: anticipation-driven perception and reinforcement-driven learning. Imitating these top-down mechanisms, together with the state-of-the-art bottom-up feed-forward algorithms, resulted in an accurate, robust, and continuously improving target recognition model. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Figure 1

Figure 1
<p>The powerful effects of context.</p> Full article ">Figure 2
<p>Schematic description of pathways in the human visual system.</p> Full article ">Figure 3
<p>Top-level schematic illustration of the pathways in the Visual Associative Predictive (VAP) model.</p> Full article ">Figure 4
<p>Schematic illustration of the VAP model attention mechanism.</p> Full article ">Figure 5
<p>Illustration of the inferior temporal cortex (ITC) mechanism in the VAP model.</p> Full article ">Figure 6
<p>Illustration of the reinforcement mechanism in the VAP model.</p> Full article ">Figure 7
<p>Schematic illustration of (<b>a</b>) the main pathways in the human visual system and (<b>b</b>) the pathways in the VAP model, using the same color code.</p> Full article ">Figure 8
<p>Comparison for all classes using the complete VAP model and using only the bottom-up pathway (VGG16).</p> Full article ">Figure 9
<p>Overall comparison between using the complete VAP model and using only the bottom-up pathway (VGG16). Precision, recall, and F1-score are shown.</p> Full article ">Figure 10
<p>Performance on video sources of the VAP model with the VGG1024 faster RCNN as the bottom-up classifier.</p> Full article ">Figure 11
<p>Performance on video sources of the VAP model with the VGG16 faster RCNN as the bottom-up classifier.</p> Full article ">Figure 12
<p>Classification results of the VAP model for bicyclists passing through the test scene. Examining the confidence level of the system for (<b>a</b>) the first and second bicyclists, (<b>b</b>) the third bicyclist, (<b>c</b>) the seventh bicyclist, and (<b>d</b>) the eight bicyclist shows gradual increase in the performance of the system.</p> Full article ">Figure 13
<p>Classification results of the bottom-up pathway (Faster Region-based Convolutional Neural Network (FRCNN) with ZF model) for bicyclists passing through the test scene. Examining the confidence level of the system for the (<b>a</b>) first and second bicyclists, (<b>b</b>) the third bicyclist, (<b>c</b>) the seventh bicyclist, and (<b>d</b>) the eight bicyclist shows constant performance. In addition, one of the people in (d) is misclassified as a plant due to the vegetation behind him.</p> Full article ">Figure 14
<p>Example of the VAP system operation. While the results of the VAP model, after watching the alley for about 30 min, resulted in high confidence recognition of the objects (<b>a</b>), using the same model without reinforcement learning resulted in failure to recognize one object and recognition of the other one with low confidence (<b>b</b>).</p> Full article ">
14 pages, 6386 KiB  
Article
Image Edge Detector with Gabor Type Filters Using a Spiking Neural Network of Biologically Inspired Neurons
by Krishnamurthy V. Vemuru
Algorithms 2020, 13(7), 165; https://doi.org/10.3390/a13070165 - 9 Jul 2020
Cited by 10 | Viewed by 4367
Abstract
We report the design of a Spiking Neural Network (SNN) edge detector with biologically inspired neurons that has a conceptual similarity with both Hodgkin-Huxley (HH) model neurons and Leaky Integrate-and-Fire (LIF) neurons. The computation of the membrane potential, which is used to determine [...] Read more.
We report the design of a Spiking Neural Network (SNN) edge detector with biologically inspired neurons that has a conceptual similarity with both Hodgkin-Huxley (HH) model neurons and Leaky Integrate-and-Fire (LIF) neurons. The computation of the membrane potential, which is used to determine the occurrence or absence of spike events, at each time step, is carried out by using the analytical solution to a simplified version of the HH neuron model. We find that the SNN based edge detector detects more edge pixels in images than those obtained by a Sobel edge detector. We designed a pipeline for image classification with a low-exposure frame simulation layer, SNN edge detection layers as pre-processing layers and a Convolutional Neural Network (CNN) as a classification module. We tested this pipeline for the task of classification with the Digits dataset, which is available in MATLAB. We find that the SNN based edge detection layer increases the image classification accuracy at lower exposure times, that is, for 1 < t < T /4, where t is the number of milliseconds in a simulated exposure frame and T is the total exposure time, with reference to a Sobel edge or Canny edge detection layer in the pipeline. These results pave the way for developing novel cognitive neuromorphic computing architectures for millisecond timescale detection and object classification applications using event or spike cameras. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Graphical abstract

Graphical abstract
Full article ">Figure 1
<p>The electrical equivalent circuit of (<b>A</b>) Hodgkin-Huxley neuron model and (<b>B</b>) Leaky Integrate-and-Fire neuron model.</p> Full article ">Figure 2
<p>Spiking Neural Network Architecture showing the neuron spike processing layers of the edge detector.</p> Full article ">Figure 3
<p>Flow chart showing the computational steps for the implementation of the Spiking Neural Network (SNN) edge detector in MATLAB with I as the input image of <math display="inline"><semantics> <mrow> <mi>m</mi> <mo>×</mo> <mi>n</mi> </mrow> </semantics></math> pixels.</p> Full article ">Figure 4
<p>Examples images along with the comparison of edges generated by SNN, Canny and Sobel edge detectors.</p> Full article ">Figure 5
<p>Comparison of edge detection methods for low exposure frames, <span class="html-italic">t</span> = 1 − 50 ms with an example image from MIT Sun Database.</p> Full article ">Figure 6
<p>Total number of edge pixels in the SNN edge map of the input image as a function of simulated exposure time t.</p> Full article ">Figure 7
<p>The process flow of the pipeline, which combines a 1 ms frame generator, SNN edge detector and a Convolutional Neural Network (CNN), for Digits classification.</p> Full article ">Figure 8
<p>The architecture of the CNN used for Digits classification.</p> Full article ">Figure 9
<p>Comparison of edge detection methods for low exposure frames, <span class="html-italic">t</span> = 1 − 40 ms with an example image from Digits dataset.</p> Full article ">Figure 10
<p>Classification accuracy as a function of number of 1 ms in the image used for edge detection.</p> Full article ">
18 pages, 767 KiB  
Article
Metric Embedding Learning on Multi-Directional Projections
by Gábor Kertész
Algorithms 2020, 13(6), 133; https://doi.org/10.3390/a13060133 - 29 May 2020
Cited by 4 | Viewed by 3311
Abstract
Image based instance recognition is a difficult problem, in some cases even for the human eye. While latest developments in computer vision—mostly driven by deep learning—have shown that high performance models for classification or categorization can be engineered, the problem of discriminating similar [...] Read more.
Image based instance recognition is a difficult problem, in some cases even for the human eye. While latest developments in computer vision—mostly driven by deep learning—have shown that high performance models for classification or categorization can be engineered, the problem of discriminating similar objects with a low number of samples remain challenging. Advances from multi-class classification are applied for object matching problems, as the feature extraction techniques are the same; nature-inspired multi-layered convolutional nets learn the representations, and the output of such a model maps them to a multidimensional encoding space. A metric based loss brings same instance embeddings close to each other. While these solutions achieve high classification performance, low efficiency is caused by memory cost of high parameter number, which is in a relationship with input image size. Upon shrinking the input, the model requires less trainable parameters, while performance decreases. This drawback is tackled by using compressed feature extraction, e.g., projections. In this paper, a multi-directional image projection transformation with fixed vector lengths (MDIPFL) is applied for one-shot recognition tasks, trained on Siamese and Triplet architectures. Results show, that MDIPFL based approach achieves decent performance, despite of the significantly lower number of parameters. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Figure 1

Figure 1
<p>The architecture of the Siamese Neural Network. The underlying backbone network is shared, the output of the structure is a similarity score.</p> Full article ">Figure 2
<p>The architecture of a triplet based embedding network. The embedding model is the same between the three samples, the output encoding is used for metric-based classification.</p> Full article ">Figure 3
<p>Samples of the Radon and MDIPFL transformations: from left to right the original images, the Radon transformations and the MDIPFL transformations, respectively. Both projection transformations show the projection profiles on range [0; <span class="html-italic">π</span>]. The original images are samples from the NIST [<a href="#B49-algorithms-13-00133" class="html-bibr">49</a>] dataset. Subfigure (<b>a</b>,<b>b</b>) show observations from the same class, subfigure (<b>c</b>) is different.</p> Full article ">Figure 4
<p>Visualization of the MDIPFL transformation. Intensity values of covered elements are summed, in the rate of coverage.</p> Full article ">Figure 5
<p>Illustration of different types of triplets. Subfigure (<b>a</b>) shows an easy negative, where <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>n</sub></span>) &gt; <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>p</sub></span>) + <span class="html-italic">m</span>, resulting in zero loss. Subfigure (<b>b</b>) illustrates a semi-hard negative, as <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>p</sub></span>) &lt; <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>n</sub></span>) &lt; <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>p</sub></span>) + <span class="html-italic">m</span>, resulting in a positive loss. On subfigure (<b>c</b>), the so-called hard negative is visualized, where <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>n</sub></span>) &lt; <span class="html-italic">d</span>(<span class="html-italic">x<sub>a</sub></span>, <span class="html-italic">x<sub>p</sub></span>).</p> Full article ">Figure 6
<p>Samples of the NIST dataset before and after content highlighting, and MDIPFL transformation. Subfigures (<b>a</b>,<b>b</b>) represent character lowercase <span class="html-italic">e</span>. Subfigure (<b>c</b>) shows a sample of lowercase <span class="html-italic">x</span>.</p> Full article ">Figure 7
<p>The evolution of loss during different phases of training. Subfigure (<b>a</b>) illustrates the measured pretraining loss for the three different inputs and models. (<b>b</b>) visualizes the measured loss on Siamese structured training, while (<b>c</b>,<b>d</b>) shows the change of triplet loss over training iterations for semi-hard and hardest negative selection, respectively. It is notable, that (<b>b</b>–<b>d</b>) use a logarithmic vertical scale for a more detailed representation.</p> Full article ">Figure 8
<p>Sample images from the Automatic Traffic Surveillance Workshop on CVPR2016 [<a href="#B66-algorithms-13-00133" class="html-bibr">66</a>]. All vehicles are recorded from two slightly different viewpoints. The dataset also contains annotation information that can be used to extract observations.</p> Full article ">Figure 9
<p>Samples of the dataset prepared using the data presented in [<a href="#B66-algorithms-13-00133" class="html-bibr">66</a>]. Vehicle images are cropped to square, and transformed using the MDIPFL projection mapping. Subfigures (<b>a</b>,<b>b</b>) represent different observations of the same instance. Subfigure (<b>c</b>) shows a sample of a different instance.</p> Full article ">Figure 10
<p>Loss over time when training ResNet-18 using the MDIPFL50 transforms of the extracted samples of the ATS-CVPR2016 dataset. Subfigure (<b>a</b>) shows the linear pretraining loss, while subfigure (<b>b</b>) visualizes the contrastive and triplet loss over iterations for different setups.</p> Full article ">Figure 11
<p>t-SNE [<a href="#B67-algorithms-13-00133" class="html-bibr">67</a>] based visualization of the 128-dimensional embedding vectors from the test dataset in two-dimensions. Subfigure (<b>a</b>) illustrates the point-cloud before training, while subfigure (<b>b</b>) visualizes it afterwards. For the transformation perplexity was set to 30 and the learning rate to 200.</p> Full article ">
14 pages, 4959 KiB  
Article
PUB-SalNet: A Pre-Trained Unsupervised Self-Aware Backpropagation Network for Biomedical Salient Segmentation
by Feiyang Chen, Ying Jiang, Xiangrui Zeng, Jing Zhang, Xin Gao and Min Xu
Algorithms 2020, 13(5), 126; https://doi.org/10.3390/a13050126 - 19 May 2020
Cited by 3 | Viewed by 5010
Abstract
Salient segmentation is a critical step in biomedical image analysis, aiming to cut out regions that are most interesting to humans. Recently, supervised methods have achieved promising results in biomedical areas, but they depend on annotated training data sets, which requires labor and [...] Read more.
Salient segmentation is a critical step in biomedical image analysis, aiming to cut out regions that are most interesting to humans. Recently, supervised methods have achieved promising results in biomedical areas, but they depend on annotated training data sets, which requires labor and proficiency in related background knowledge. In contrast, unsupervised learning makes data-driven decisions by obtaining insights directly from the data themselves. In this paper, we propose a completely unsupervised self-aware network based on pre-training and attentional backpropagation for biomedical salient segmentation, named as PUB-SalNet. Firstly, we aggregate a new biomedical data set from several simulated Cellular Electron Cryo-Tomography (CECT) data sets featuring rich salient objects, different SNR settings, and various resolutions, which is called SalSeg-CECT. Based on the SalSeg-CECT data set, we then pre-train a model specially designed for biomedical tasks as a backbone module to initialize network parameters. Next, we present a U-SalNet network to learn to selectively attend to salient objects. It includes two types of attention modules to facilitate learning saliency through global contrast and local similarity. Lastly, we jointly refine the salient regions together with feature representations from U-SalNet, with the parameters updated by self-aware attentional backpropagation. We apply PUB-SalNet for analysis of 2D simulated and real images and achieve state-of-the-art performance on simulated biomedical data sets. Furthermore, our proposed PUB-SalNet can be easily extended to 3D images. The experimental results on the 2d and 3d data sets also demonstrate the generalization ability and robustness of our method. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Figure 1

Figure 1
<p>An unsupervised example of semantic segmentation and salient segmentation on CECT images.</p> Full article ">Figure 2
<p>The overview framework of our proposed method. The processing pipeline consists of three main steps: (<b>a</b>) pre-training on the SalSeg-CECT data set; (<b>b</b>) prediction using the U-SalNet model; (<b>c</b>) unsupervised attentional backpropagation iterating on single images. The <span style="color:blue">Enc</span> and <span style="color:yellow">Dec</span> stand for the encoder and decoder. The <span style="color:red">⨂</span>, <span style="color:green">⨂</span>, and <span style="color:blue">⨂</span> denote the global attention mechanism, local attention mechanism and convolutional decoding, respectively.</p> Full article ">Figure 3
<p>The architecture of our U-SalNet model. (<b>a</b>) Global Attention and (<b>b</b>) Local Attention corresponds to <span style="color:red">⨂</span> and <span style="color:green">⨂</span> from <a href="#algorithms-13-00126-f002" class="html-fig">Figure 2</a>b, respectively. GA and LA stand for Global Attention and Local Attention. Conv means the convolution operation. ⨁ stands for weighted summation over the feature map.</p> Full article ">Figure 4
<p>Qualitative visual results of ten unsupervised methods on the simulated biomedical data set with SNR = 0.5 and 1.5. GT stands for ground truth images, PUB is PUB-SalNet, and the other nine methods are referenced in <a href="#algorithms-13-00126-t001" class="html-table">Table 1</a>.</p> Full article ">Figure 5
<p>Visualization of 3D salient segmentation by PUB-SalNet on a 3D subvolume of size <math display="inline"><semantics> <mrow> <mn>64</mn> <mo>×</mo> <mn>64</mn> <mo>×</mo> <mn>64</mn> </mrow> </semantics></math> from the CECT test set. The pictures are obtained using UCSF Chimera, which displays the isosurface of the four corresponding 3D images; (<b>a</b>) is the original image with a threshold of 0 (<b>b</b>) is the ground truth of macro-molecular structures (<b>c</b>) is our prediction (<b>d</b>) demonstrates that the predicted salient region greatly overlaps with the ground truth macro-molecular structure.</p> Full article ">Figure 6
<p>Case study of 2D salient segmentation by PUB-SalNet and the B module on the ISBI Challenge [<a href="#B27-algorithms-13-00126" class="html-bibr">27</a>].</p> Full article ">
16 pages, 11276 KiB  
Article
Oil Spill Monitoring of Shipborne Radar Image Features Using SVM and Local Adaptive Threshold
by Jin Xu, Haixia Wang, Can Cui, Baigang Zhao and Bo Li
Algorithms 2020, 13(3), 69; https://doi.org/10.3390/a13030069 - 21 Mar 2020
Cited by 29 | Viewed by 4357
Abstract
In the case of marine accidents, monitoring marine oil spills can provide an important basis for identifying liabilities and assessing the damage. Shipborne radar can ensure large-scale, real-time monitoring, in all weather, with high-resolution. It therefore has the potential for broad applications in [...] Read more.
In the case of marine accidents, monitoring marine oil spills can provide an important basis for identifying liabilities and assessing the damage. Shipborne radar can ensure large-scale, real-time monitoring, in all weather, with high-resolution. It therefore has the potential for broad applications in oil spill monitoring. Considering the original gray-scale image from the shipborne radar acquired in the case of the Dalian 7.16 oil spill accident, a complete oil spill detection method is proposed. Firstly, the co-frequency interferences and speckles in the original image are eliminated by preprocessing. Secondly, the wave information is classified using a support vector machine (SVM), and the effective wave monitoring area is generated according to the gray distribution matrix. Finally, oil spills are detected by a local adaptive threshold and displayed on an electronic chart based on geographic information system (GIS). The results show that the SVM can extract the effective wave information from the original shipborne radar image, and the local adaptive threshold method has strong applicability for oil film segmentation. This method can provide a technical basis for real-time cleaning and liability determination in oil spill accidents. Full article
(This article belongs to the Special Issue Bio-Inspired Algorithms for Image Processing)
►▼ Show Figures

Figure 1

Figure 1
<p>Original radar images, where (<b>a</b>–<b>d</b>) were acquired at 23:19:58, 23:20:27, 23:21:03, and 23:21:09, respectively.</p> Full article ">Figure 2
<p>Clean-up mission. The red line indicates the departure route and the green line indicates the return route.</p> Full article ">Figure 3
<p>Hardware architecture showing (<b>a</b>) the installation position of the radar antenna and the (<b>b</b>) radar data acquisition system.</p> Full article ">Figure 4
<p>Image generation mechanism.</p> Full article ">Figure 5
<p>Flow of data preprocessing.</p> Full article ">Figure 6
<p>Result of data preprocessing showing (<b>a</b>) convolution of the Laplace operator, (<b>b</b>) threshold segmentation, <span class="html-italic">T</span> = 128, (<b>c</b>) in the 1 × 7 window centered on the highlighted pixels of the original image, with two nearest non-highlighted points selected to replace the highlighted pixel with the mean value, (<b>d</b>) segmentation of isolated targets, <span class="html-italic">T<sub>area</sub></span> = 200, and (<b>e</b>) the window of median filtering, which is 21×21.</p> Full article ">Figure 7
<p>Flow of oil film classification. SVM—support vector machine.</p> Full article ">Figure 8
<p>Optimal hyperplane for binary classification.</p> Full article ">Figure 9
<p>Classification of <a href="#algorithms-13-00069-f006" class="html-fig">Figure 6</a>e. Long-range shoreline echoes have been deleted.</p> Full article ">Figure 10
<p>Determination of effective wave area showing (<b>a</b>) gray gradient matrix of <a href="#algorithms-13-00069-f009" class="html-fig">Figure 9</a>, (<b>b</b>) the mask of gray threshold “20”, and (<b>c</b>) the final, pre-analyzed area.</p> Full article ">Figure 11
<p>Results of the local adaptive threshold with (<b>a</b>) initial identification and (<b>b</b>) fine identification.</p> Full article ">Figure 12
<p>Oil film classification results, where (<b>a</b>–<b>d</b>) are results of <a href="#algorithms-13-00069-f001" class="html-fig">Figure 1</a>a–d, respectively.</p> Full article ">Figure 13
<p>Location of the oil spill shown in the electronic chart.</p> Full article ">Figure 14
<p>Visible light images with oil films were acquired during the day. (<b>a</b>,<b>b</b>) are airborne and shipborne visible light images, respectively.</p> Full article ">Figure 15
<p>Thermal infrared images with oil films were acquired during the night.</p> Full article ">Figure 16
<p>Scattering mechanism of shipborne radar electromagnetic waves.</p> Full article ">Figure 17
<p>Oil spill imaging mechanism of the shipborne radar images. Oil film reduces the backscatter of the shipborne radar image.</p> Full article ">Figure 18
<p>Comparison with other local adaptive thresholds of <a href="#algorithms-13-00069-f001" class="html-fig">Figure 1</a>b showing (<b>a</b>) Sauvola’s method, (<b>b</b>) Bernsen’s method, and (<b>c</b>) the Otsu method.</p> Full article ">Figure 19
<p>Comparison with other oil film classifications of <a href="#algorithms-13-00069-f001" class="html-fig">Figure 1</a>b showing (<b>a</b>) Liu’s method, (<b>b</b>) Zhu’s method (<span class="html-italic">T</span><sub>gray</sub> = 55), and (<b>c</b>) Xu’s method (<span class="html-italic">T</span><sub>gray</sub> = 100, <span class="html-italic">Tarea</span> = 30).</p> Full article ">
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:
 
Displaying articles 1-7
Back to TopTop