Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563413
Marcin Kociolek, P. Bajcsy, M. Brady, Antonio Cardone
A novel interpolation-based model for the computation of the Gray Level Co-occurrence Matrix (GLCM) is presented. The model enables GLCM computation for any real-valued angles and offsets, as opposed to the traditional, lattice-based model. A texture directionality estimation algorithm is defined using the GLCM-derived correlation feature. The robustness of the algorithm with respect to image blur and additive Gaussian noise is evaluated. It is concluded that directionality estimation is robust to image blur and low noise levels. For high noise levels, the mean error increases but remains bounded. The performance of the directionality estimation algorithm is illustrated on fluorescence microscopy images of fibroblast cells. The algorithm was implemented in C++ and the source code is available in an openly accessible repository.
{"title":"Interpolation-Based Gray-Level Co-Occurrence Matrix Computation for Texture Directionality Estimation","authors":"Marcin Kociolek, P. Bajcsy, M. Brady, Antonio Cardone","doi":"10.23919/SPA.2018.8563413","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563413","url":null,"abstract":"A novel interpolation-based model for the computation of the Gray Level Co-occurrence Matrix (GLCM) is presented. The model enables GLCM computation for any real-valued angles and offsets, as opposed to the traditional, lattice-based model. A texture directionality estimation algorithm is defined using the GLCM-derived correlation feature. The robustness of the algorithm with respect to image blur and additive Gaussian noise is evaluated. It is concluded that directionality estimation is robust to image blur and low noise levels. For high noise levels, the mean error increases but remains bounded. The performance of the directionality estimation algorithm is illustrated on fluorescence microscopy images of fibroblast cells. The algorithm was implemented in C++ and the source code is available in an openly accessible repository.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126031749","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563368
Sebastian Cygert, A. Czyżewski
Vehicle detection in video from a miniature stationary closed-circuit television (CCTV) camera is discussed in the paper. The camera provides one of components of the intelligent road sign developed in the project concerning the traffic control with the use of autonomous devices being developed. Modern Convolutional Neural Network (CNN) based detectors need big data input, usually demanding their manual labeling. In the presented research approach the weakly-supervised learning paradigm is used for the training of a CNN based detector employing labels obtained automatically through an application of video background subtraction algorithm. The proposed method is evaluated on GRAM-RTM dataset and a CNN fine-tuned with labels from the background subtraction algorithm. Even though obtained representation in the form of labels may include many false positives and negatives, a reliable vehicle detector was trained employing them. The results are presented showing that such a method can be applied to traffic surveillance systems.
{"title":"Vehicle detector training with labels derived from background subtraction algorithms in video surveillance","authors":"Sebastian Cygert, A. Czyżewski","doi":"10.23919/SPA.2018.8563368","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563368","url":null,"abstract":"Vehicle detection in video from a miniature stationary closed-circuit television (CCTV) camera is discussed in the paper. The camera provides one of components of the intelligent road sign developed in the project concerning the traffic control with the use of autonomous devices being developed. Modern Convolutional Neural Network (CNN) based detectors need big data input, usually demanding their manual labeling. In the presented research approach the weakly-supervised learning paradigm is used for the training of a CNN based detector employing labels obtained automatically through an application of video background subtraction algorithm. The proposed method is evaluated on GRAM-RTM dataset and a CNN fine-tuned with labels from the background subtraction algorithm. Even though obtained representation in the form of labels may include many false positives and negatives, a reliable vehicle detector was trained employing them. The results are presented showing that such a method can be applied to traffic surveillance systems.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114078980","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563389
P. Kłosowski
The article presents an example of practical application of deep learning methods for language processing and modelling. Development of statistical language models helps to predict a sequence of recognized words and phonemes, and can be used for improving speech processing and speech recognition. However, currently the field of language modelling is shifting from statistical language modelling methods to neural networks and deep learning methods. Therefore, one of the methods of effective language modelling with the use of deep learning techniques is presented in this paper. Presented results concerns the modelling of the Polish language but the achieved research results and conclusions can also be applied to language modelling application for other languages.
{"title":"Deep Learning for Natural Language Processing and Language Modelling","authors":"P. Kłosowski","doi":"10.23919/SPA.2018.8563389","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563389","url":null,"abstract":"The article presents an example of practical application of deep learning methods for language processing and modelling. Development of statistical language models helps to predict a sequence of recognized words and phonemes, and can be used for improving speech processing and speech recognition. However, currently the field of language modelling is shifting from statistical language modelling methods to neural networks and deep learning methods. Therefore, one of the methods of effective language modelling with the use of deep learning techniques is presented in this paper. Presented results concerns the modelling of the Polish language but the achieved research results and conclusions can also be applied to language modelling application for other languages.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122430839","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563269
A. Borowicz
Independent component analysis (ICA) is a popular technique for demixing multi-sensor data. In many approaches to the ICA, signals are decorrelated by whitening data and then by rotating the result. In this paper, we introduce a four-unit, symmetric algorithm, based on quaternionic factorization of rotation matrix. It makes use an isomorphism between quaternions and $4times 4$ orthogonal matrices. Unlike conventional techniques based on Jacobi decomposition, our method exploits 4D rotations and uses negentropy approximation as a contrast function. Compared to the widely used, symmetric FastICA algorithm, the proposed method offers a better separation quality in a presence of multiple Gaussian sources.
{"title":"On Using Quaternionic Rotations for Indpendent Component Analysis","authors":"A. Borowicz","doi":"10.23919/SPA.2018.8563269","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563269","url":null,"abstract":"Independent component analysis (ICA) is a popular technique for demixing multi-sensor data. In many approaches to the ICA, signals are decorrelated by whitening data and then by rotating the result. In this paper, we introduce a four-unit, symmetric algorithm, based on quaternionic factorization of rotation matrix. It makes use an isomorphism between quaternions and $4times 4$ orthogonal matrices. Unlike conventional techniques based on Jacobi decomposition, our method exploits 4D rotations and uses negentropy approximation as a contrast function. Compared to the widely used, symmetric FastICA algorithm, the proposed method offers a better separation quality in a presence of multiple Gaussian sources.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122884356","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563312
H. Orimoto, A. Ikuta
Numerous noise suppression methods for speech signals have been developed up to now. In this paper, a new method to suppress noise in speech signals is proposed by use of an extension type Unscented Kalman filter (UKF). A method considering non-Gaussian noise is proposed theoretically by introducing an expansion expression of Bayes' theorem and considering nonlinear correlation information between the speech signal and the observation data. Specifically, by selecting appropriately the sample points and the weight coefficients, an estimation algorithm of the speech signal for nonliner system is derived on the basis of conditional probability distribution. Moreover, expansion coefficients in the estimation algorithm are realized by considering the higher order correlation information. Improvement for the precise estimation is expected by considering non-Gaussian property. The effectiveness of the proposed method is confirmed by applying it to speech signals contaminated by noises.
{"title":"Noise Cancellation Method for Speech Signal by Using an Extension Type UKF","authors":"H. Orimoto, A. Ikuta","doi":"10.23919/SPA.2018.8563312","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563312","url":null,"abstract":"Numerous noise suppression methods for speech signals have been developed up to now. In this paper, a new method to suppress noise in speech signals is proposed by use of an extension type Unscented Kalman filter (UKF). A method considering non-Gaussian noise is proposed theoretically by introducing an expansion expression of Bayes' theorem and considering nonlinear correlation information between the speech signal and the observation data. Specifically, by selecting appropriately the sample points and the weight coefficients, an estimation algorithm of the speech signal for nonliner system is derived on the basis of conditional probability distribution. Moreover, expansion coefficients in the estimation algorithm are realized by considering the higher order correlation information. Improvement for the precise estimation is expected by considering non-Gaussian property. The effectiveness of the proposed method is confirmed by applying it to speech signals contaminated by noises.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128009348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563406
Karolina Marciniuk, M. Szczodrak, A. Czyżewski
Assessment of road traffic parameters for the developed intelligent speed limit setting decision system constitutes the subject addressed in the paper. Current traffic conditions providing vital data source for the calculation of the locally fitted speed limits are assessed employing an economical embedded platform placed at the roadside. The use of the developed platform employing a low-powered processing unit with a set of microphones, an accelerometer and some other sensors, for the estimation of the essential road traffic parameters is presented in the paper. Acoustical signal processing-based vehicle counting attempts were made, and an acceleration sensor was used in order to detect the heavy vehicles pass-bys. Obtained results based on the measurements were discussed in the paper. Evaluation of the proposed methods is provided.
{"title":"An application of acoustic sensors for the monitoring of road traffic","authors":"Karolina Marciniuk, M. Szczodrak, A. Czyżewski","doi":"10.23919/SPA.2018.8563406","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563406","url":null,"abstract":"Assessment of road traffic parameters for the developed intelligent speed limit setting decision system constitutes the subject addressed in the paper. Current traffic conditions providing vital data source for the calculation of the locally fitted speed limits are assessed employing an economical embedded platform placed at the roadside. The use of the developed platform employing a low-powered processing unit with a set of microphones, an accelerometer and some other sensors, for the estimation of the essential road traffic parameters is presented in the paper. Acoustical signal processing-based vehicle counting attempts were made, and an acceleration sensor was used in order to detect the heavy vehicles pass-bys. Obtained results based on the measurements were discussed in the paper. Evaluation of the proposed methods is provided.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134237661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563388
C. Vinhais, M. Kociński, A. Materka
Accurate modeling of the human vascular tree from 3D computed tomography (CTA) or magnetic resonance (MRA) angiograms is required for visualization, diagnosis of vascular diseases, and computational fluid dynamic (CFD) blood flow simulations. This work describes an automated algorithm for constructing the polygonal mesh of blood vessels from such images. Each vascular segment is modeled as a tubular object, and a thin plate spline transform is used to generate the corresponding surface from its centerline-radius representation. A novel approach for generating the polygonal mesh of bifurcating vessels based on conformal mapping is presented. A mathematical description of the methodology is also provided. The model is improved by computing local intensity features with subvoxel accuracy, to slightly deform the mesh of the vascular tree for fine-tuning. The proposed algorithm was successfully tested on a 3D synthetic image containing randomly generated vascular branches. Experiment results, confirmed by real-world Time of Flight MRA, demonstrate that our methodology is consistent and capable of generating high quality triangulated meshes of vascular trees, suitable for further CFD simulations. Compared to common techniques, conformal mapping proved to be a simple and effective mathematical approach for polygonal mesh modeling of bifurcating vessels.
{"title":"Centerline-Radius Polygonal-Mesh Modeling of Bifurcated Blood Vessels in 3D Images using Conformal Mapping","authors":"C. Vinhais, M. Kociński, A. Materka","doi":"10.23919/SPA.2018.8563388","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563388","url":null,"abstract":"Accurate modeling of the human vascular tree from 3D computed tomography (CTA) or magnetic resonance (MRA) angiograms is required for visualization, diagnosis of vascular diseases, and computational fluid dynamic (CFD) blood flow simulations. This work describes an automated algorithm for constructing the polygonal mesh of blood vessels from such images. Each vascular segment is modeled as a tubular object, and a thin plate spline transform is used to generate the corresponding surface from its centerline-radius representation. A novel approach for generating the polygonal mesh of bifurcating vessels based on conformal mapping is presented. A mathematical description of the methodology is also provided. The model is improved by computing local intensity features with subvoxel accuracy, to slightly deform the mesh of the vascular tree for fine-tuning. The proposed algorithm was successfully tested on a 3D synthetic image containing randomly generated vascular branches. Experiment results, confirmed by real-world Time of Flight MRA, demonstrate that our methodology is consistent and capable of generating high quality triangulated meshes of vascular trees, suitable for further CFD simulations. Compared to common techniques, conformal mapping proved to be a simple and effective mathematical approach for polygonal mesh modeling of bifurcating vessels.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114152817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563412
M. Strzelecki, A. Klepaczko, Martyna Muszelska, E. Eikefjord, J. Rørvik, A. Lundervold
The dynamic contrast-enhanced magnetic resonance imaging is a diagnostic method directed at estimation of renal performance. Analysis of the image intensity time-courses in the renal cortex and parenchyma enables quantification of the kidney filtration characteristics. A standard approach used for that purpose involves fitting a pharmacokinetic model to image data and optimizing a set of model parameters. It is essentially a multi-objective and non-linear optimization problem. Standard methods applied in such scenarios include nonlinear least-squares (NLS) algorithms, such as Levenberg-Marquardt or Trust Region Reflective methods. The major disadvantage of these classical approaches is the requirement for determining the starting point of the optimization, whose final result is a local minimum of the objective function. On the contrary, artificial neural networks (ANN) are trained based on a large range of parameter combinations, potentially covering whole solution space. Thus, they appear particularly useful in fitting complex, non-linear, multi-parametric relationships to the observed noisy data and offer greater ability to detect all possible interactions between predictor variables without the need for explicit statistical formulation. In this paper we compare the ANN and NLS approaches in application to measuring perfusion based on DCE-MR images. The experiments performed on a dataset containing 10 dynamic image series collected for 5 healthy volunteers proved superior performance of the neural networks over classical methods in terms of quantifying true perfusion parameters, robustness to noise and varying imaging conditions.
动态对比增强磁共振成像是一种诊断方法，旨在估计肾脏的表现。分析肾皮质和实质的图像强度时程可以量化肾脏滤过特性。用于此目的的标准方法包括将药代动力学模型拟合到图像数据并优化一组模型参数。它本质上是一个多目标非线性优化问题。在这种情况下应用的标准方法包括非线性最小二乘(NLS)算法，如Levenberg-Marquardt或Trust Region Reflective方法。这些经典方法的主要缺点是需要确定优化的起始点，其最终结果是目标函数的局部最小值。相反，人工神经网络(ANN)是基于大范围的参数组合来训练的，有可能覆盖整个解空间。因此，它们在拟合观察到的噪声数据的复杂、非线性、多参数关系方面显得特别有用，并且在不需要显式统计公式的情况下，提供了更大的能力来检测预测变量之间所有可能的相互作用。在本文中，我们比较了ANN和NLS方法在DCE-MR图像灌注测量中的应用。在包含5名健康志愿者的10个动态图像序列数据集上进行的实验证明，神经网络在量化真实灌注参数、对噪声的鲁棒性和不同成像条件方面优于经典方法。
{"title":"An artificial neural network for GFR estimation in the DCE-MRI studies of the kidneys","authors":"M. Strzelecki, A. Klepaczko, Martyna Muszelska, E. Eikefjord, J. Rørvik, A. Lundervold","doi":"10.23919/SPA.2018.8563412","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563412","url":null,"abstract":"The dynamic contrast-enhanced magnetic resonance imaging is a diagnostic method directed at estimation of renal performance. Analysis of the image intensity time-courses in the renal cortex and parenchyma enables quantification of the kidney filtration characteristics. A standard approach used for that purpose involves fitting a pharmacokinetic model to image data and optimizing a set of model parameters. It is essentially a multi-objective and non-linear optimization problem. Standard methods applied in such scenarios include nonlinear least-squares (NLS) algorithms, such as Levenberg-Marquardt or Trust Region Reflective methods. The major disadvantage of these classical approaches is the requirement for determining the starting point of the optimization, whose final result is a local minimum of the objective function. On the contrary, artificial neural networks (ANN) are trained based on a large range of parameter combinations, potentially covering whole solution space. Thus, they appear particularly useful in fitting complex, non-linear, multi-parametric relationships to the observed noisy data and offer greater ability to detect all possible interactions between predictor variables without the need for explicit statistical formulation. In this paper we compare the ANN and NLS approaches in application to measuring perfusion based on DCE-MR images. The experiments performed on a dataset containing 10 dynamic image series collected for 5 healthy volunteers proved superior performance of the neural networks over classical methods in terms of quantifying true perfusion parameters, robustness to noise and varying imaging conditions.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116589638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563411
Jakub Jurek, M. Kociński, A. Materka, Are Losnegård, L. Reisæter, O. Halvorsen, C. Beisland, J. Rørvik, A. Lundervold
T2-weighted magnetic resonance images (T2W MRI) of prostate cancer are usually acquired with a large slice thickness compared to in-plane voxel dimensions and to the minimal significant malignant prostate tumour size. This causes a negative partial volume effect, decreasing the precision of tumour volumetry and complicating 3D texture analysis of the images. At the same time, three orthogonal, anisotropic acquisitions with overlapping fields of view are often acquired to allow insight into the prostate from different anatomical planes. It is desirable to reconstruct an isotropic prostate T2W image, using the 3 orthogonal volumes computationally, instead of directly acquiring a high-resolution MR image, which typically requires elongated scanning time, with higher cost, less patient comfort and lower signal-to-noise ratio. In our previous work, we followed the above rationale applying a Markov-Random-Field(MRF)-based combination of 3 orthogonal T2W images of the prostate. Our initial results were, however, biased by the quality of input orthogonal images. These were first preprocessed using spline interpolation to yield the same voxel dimensions and later registered. In this paper, we apply a dictionary learning approach to interpolation in order to increase the resolution of a coronal T2W MRI image. We compose a low-resolution dictionary from the original axial image, calculate its sparse representation by Orthogonal Matching Pursuit and finally derive the high-resolution dictionary to improve the original coronal image. We assess the improvement in visual image quality as satisfying and propose further studies.
{"title":"Dictionary-based through-plane interpolation of prostate cancer T2-weighted MR images","authors":"Jakub Jurek, M. Kociński, A. Materka, Are Losnegård, L. Reisæter, O. Halvorsen, C. Beisland, J. Rørvik, A. Lundervold","doi":"10.23919/SPA.2018.8563411","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563411","url":null,"abstract":"T2-weighted magnetic resonance images (T2W MRI) of prostate cancer are usually acquired with a large slice thickness compared to in-plane voxel dimensions and to the minimal significant malignant prostate tumour size. This causes a negative partial volume effect, decreasing the precision of tumour volumetry and complicating 3D texture analysis of the images. At the same time, three orthogonal, anisotropic acquisitions with overlapping fields of view are often acquired to allow insight into the prostate from different anatomical planes. It is desirable to reconstruct an isotropic prostate T2W image, using the 3 orthogonal volumes computationally, instead of directly acquiring a high-resolution MR image, which typically requires elongated scanning time, with higher cost, less patient comfort and lower signal-to-noise ratio. In our previous work, we followed the above rationale applying a Markov-Random-Field(MRF)-based combination of 3 orthogonal T2W images of the prostate. Our initial results were, however, biased by the quality of input orthogonal images. These were first preprocessed using spline interpolation to yield the same voxel dimensions and later registered. In this paper, we apply a dictionary learning approach to interpolation in order to increase the resolution of a coronal T2W MRI image. We compose a low-resolution dictionary from the original axial image, calculate its sparse representation by Orthogonal Matching Pursuit and finally derive the high-resolution dictionary to improve the original coronal image. We assess the improvement in visual image quality as satisfying and propose further studies.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129950552","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/SPA.2018.8563422
J. Bednarek, K. Piaskowski, Michał Bednarek
Semantic Segmentation is one of the visual tasks that gained the significant boost in performance in recent years due to the popularization of Convolutional Neural Networks (CNNs). In this paper, we addressed the problem of losing information while changing the size of input images during training neural models. Moreover, our method of downsampling and upsampling could be easily injected into current autoencoder models. We show that without any significant changes in a model architecture it is possible to noticeably improve IoU metric. On popular Cityscapes benchmark, our model is achieving almost 2.5% boost in the accuracy of segmentation in comparison to the widely known ERF model. Additionally, to the ability to real-time usages, we run our network on GPU comparable to NVIDIA Jetson Tx2, what let us implement our algorithm in autonomous vehicles.
{"title":"Methods of Enriching The Flow of Information in The Real-Time Semantic Segmentation Using Deep Neural Networks","authors":"J. Bednarek, K. Piaskowski, Michał Bednarek","doi":"10.23919/SPA.2018.8563422","DOIUrl":"https://doi.org/10.23919/SPA.2018.8563422","url":null,"abstract":"Semantic Segmentation is one of the visual tasks that gained the significant boost in performance in recent years due to the popularization of Convolutional Neural Networks (CNNs). In this paper, we addressed the problem of losing information while changing the size of input images during training neural models. Moreover, our method of downsampling and upsampling could be easily injected into current autoencoder models. We show that without any significant changes in a model architecture it is possible to noticeably improve IoU metric. On popular Cityscapes benchmark, our model is achieving almost 2.5% boost in the accuracy of segmentation in comparison to the widely known ERF model. Additionally, to the ability to real-time usages, we run our network on GPU comparable to NVIDIA Jetson Tx2, what let us implement our algorithm in autonomous vehicles.","PeriodicalId":265587,"journal":{"name":"2018 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126772052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：