scholarly journals Computational Geometry Computation and KNN Segmentation in ITK

2006 ◽  
Author(s):  
Ruben Cardenes ◽  
Manuel Rene Sanchez ◽  
Juan Ruiz-Alzola
Keyword(s):  
Computational Geometry ◽  
Nearest Neighbor ◽  
Brain Mri ◽  
Voronoi Diagrams ◽  
K Nearest Neighbor ◽  
Fast Marching ◽  
Image Objects ◽  
Supervised Segmentation ◽  
And Performance ◽  
2D And 3D

This work describes the implementation of computational geometry algorithms developed within the Insight Toolkit (ITK): Distance Transform (DT), Voronoi diagrams, k Nearest Neighbor (kNN) transform, and finally a K Nearest Neighbor classifier for multichannel data, that is used for supervised segmentation. We have tested this algorithm for 2D and 3D medical datasets, and the results are excellent in terms of accuracy and performance. One of the strongest points of the algorithms described here is that they can be used for many other applications, because they are based on the ordered propagation paradigm. This idea consists in actually not raster scan the image but rather in start from the image objects and propagate them until the image is totally filled. This has been demonstrated to be a good approach in many algorithms as for example, computation of Distance Transforms, Voronoi Diagrams, Fast Marching, skeletons computation, etc. We show here that these algorithms have low computational complexity and it provides excellent results for clinical applications as the segmentation of brain MRI.

Processing Continuous k Nearest Neighbor Queries in Obstructed Space with Voronoi Diagrams

2020 ◽  
Vol 7 (2) ◽  
pp. 1-27
Author(s):  
Huaijie Zhu ◽  
Xiaochun Yang ◽  
Bin Wang ◽  
Wang-Chien Lee ◽  
Jian Yin ◽  
...  
Keyword(s):  
Nearest Neighbor ◽  
Voronoi Diagrams ◽  
K Nearest Neighbor ◽  
Nearest Neighbor Queries

scholarly journals Common Sport Movement Recognition from Wearable Inertial Sensor

2020 ◽  
Vol 8 (5) ◽  
pp. 1285-1292
Keyword(s):  
Nearest Neighbor ◽  
Inertial Sensor ◽  
Window Size ◽  
Evaluation Process ◽  
Support Vector ◽  
K Nearest Neighbor ◽  
Total Energy Consumption ◽  
Movement Recognition ◽  
The Common ◽  
And Performance

Common sport movements are the fundamental movements in all kind of sports. There are lots of researches done on classifying sports movements but very few are focused on common sport movement which is the focus of this project. The main aim is to develop an automated algorithm that can detect the common sport movements into walking based and jumping based movement from the wearable inertial sensor. The inertial sensor signals obtained from ten subjects were processed and grouped into walking-based and jumping-based movements. Time-domain features were extracted from the signals. Finally, the classification and performance evaluation process is done by using three different classification models (Support Vector Machine (SVM), k Nearest Neighbor (k-NN) and Decision Tree) with fixed window size of 1.28 seconds at the first stage. At the second stage, the best model from the first stage was used to determine the best window size in extracting the features that represent the walking and jumping based movement. As a result, SVM algorithm with window size of 2 seconds produced the highest overall accuracy of 95.4 % which proved to be the best classification algorithm to classify the common sport movements into walking-based and jumping-based movements. It is hoped that the outcome from this project can be used as a part of developing the overall automated sport movement recognition which is useful for the analyst, coach or player to analyse the performance of the player as well as predicting total energy consumption in preventing the injury among the player

scholarly journals Cloud-based architecture for face identification with deep learning using convolutional neural network

2021 ◽  
Vol 23 (2) ◽  
pp. 811
Author(s):  
Aditya Herlambang ◽  
Putu Wira Buana ◽  
I Nyoman Piarsa
Keyword(s):  
Neural Network ◽  
Deep Learning ◽  
Convolutional Neural Network ◽  
Nearest Neighbor ◽  
Face Identification ◽  
K Nearest Neighbor ◽  
Non Invasive ◽  
The Face ◽  
And Performance ◽  
University Of Manchester

The use of a face as a biometric to identify a person in order to keep the system safe from an unauthorized person has advantages over other biometric characteristics. The face as a biometric has more structure and a wider area than other biometrics, while can be retrieved in a non-invasive manner. We proposed a cloud-based architecture for face identification with deep learning using convolutional neural network. Face identification in this study used a cloud-based engine with four stages, namely face detection with histogram of oriented gradients (HOG), image enhancement, feature extraction using convolutional neural network, and classification using k-nearest neighbor (KNN), SVM, as well as random forest algorithm. This study conducted a classification experiment with cloud-based architecture using three different datasets, namely Faces94, Faces96 and University of Manchester Institute of Science and Technology (UMIST) face dataset. The results from this study are with the proposed cloud-based architecture, the best accuracy is obtained by KNN algorithm with an accuracy of 99% on Faces94 dataset, 99% accuracy on Faces96 dataset, 97% on UMIST face dataset, and performance of the three algorithms decreased in UMIST face dataset with facial variations from various angles from left to right profile.

scholarly journals Postal Envelope Segmentation using Learning-Based Approach

2008 ◽  
Vol 11 (2) ◽  
Author(s):  
Horacio A Legal-Ayala ◽  
Jacques Facon ◽  
Benjamín Barán
Keyword(s):  
Nearest Neighbor ◽  
Input Image ◽  
Success Rates ◽  
K Nearest Neighbor ◽  
Postal Address ◽  
New Approach ◽  
Complex Envelope ◽  
Accuracy Measure ◽  
And Performance ◽  
The Moment

This paper presents a learning-based approach to segment postal address blocks where the learning step uses only one pair of images (a sample image and its ideal segmented solution). First, this approach learns the available knowledge among pixels (each gray level) in an input image and its corresponding output in the ideal segmented solution. A classification array is generated which is re-utilized during the segmentation of new images. Features are extracted and updated by means of an adaptive square neighborhood. At the moment of new image segmentation, the submitted images are segmented by means of a k-Nearest Neighbor (k-NN) algorithm that seeks, for each pixel, the best solution in the classification array. Tests on a database of 200 complex envelope images were performed and a pixel to pixel accuracy measure validates the new approach. Results compared to other approaches for the same database show the efficiency and performance of the proposed learning-based approach. Success rates achieved for address block, stamps, rubber stamps and noise suggest that the features used in the proposed approach improves the segmentation results.

scholarly journals Modification of KNN Algorithm

2019 ◽  
Vol 8 (11) ◽  
pp. 24869-24877 ◽  
Author(s):  
Shubham Pandey ◽  
Vivek Sharma ◽  
Garima Agrawal
Keyword(s):  
Prior Knowledge ◽  
Nearest Neighbor ◽  
Main Idea ◽  
Classification Algorithm ◽  
Training Data ◽  
Data Sets ◽  
Classification Methods ◽  
K Nearest Neighbor ◽  
Knn Classification ◽  
And Performance

K-Nearest Neighbor (KNN) classification is one of the most fundamental and simple classification methods. It is among the most frequently used classification algorithm in the case when there is little or no prior knowledge about the distribution of the data. In this paper a modification is taken to improve the performance of KNN. The main idea of KNN is to use a set of robust neighbors in the training data. This modified KNN proposed in this paper is better from traditional KNN in both terms: robustness and performance. Inspired from the traditional KNN algorithm, the main idea is to classify an input query according to the most frequent tag in set of neighbor tags with the say of the tag closest to the new tuple being the highest. Proposed Modified KNN can be considered a kind of weighted KNN so that the query label is approximated by weighting the neighbors of the query. The procedure computes the frequencies of the same labeled neighbors to the total number of neighbors with value associated with each label multiplied by a factor which is inversely proportional to  the distance between new tuple and neighbours. The proposed method is evaluated on a variety of several standard UCI data sets. Experiments show the significant improvement in the performance of KNN method.

scholarly journals Estimasi Tujuan Penumpang Menggunakan Predictive Model dengan Data Smart Card

2019 ◽  
Vol 2 (2) ◽  
pp. 99-110
Author(s):  
Ekky Alam ◽  
Inkreswari Hardini ◽  
Goklas Panjaitan ◽  
Sita Rosida
Keyword(s):  
Predictive Model ◽  
Public Transportation ◽  
Urban Areas ◽  
Nearest Neighbor ◽  
Performance Indicator ◽  
K Nearest Neighbor ◽  
Fundamental Information ◽  
And Performance ◽  
Public Transportation Planning

Bus Rapid Transit (BRT) is one of the main choices of public transportation that supports mobility of Jakarta community. As one of the main choices of public transportation, BRT should provide good service and always improve its performance. Needs for moving or mobility will cause a problem if the moving itself is heading at the same area and at the same time. That will cause some problems which are often faced in urban areas such as traffic and delay. To overcome those problems there needs to be a strategy to build good public transportation planning, besides need to know individual travel patterns to overcome problems and improve BRT service. In case to realize those plans needs to be built origin-destination (O-D) matrix. O-D matrix is a matrix that each cell is an amount of trip from the source(row) to the destination (column). O-D matrix is beneficial for analysis, design and public transportation management. O-D matrix also provides useful information like amount of trip between 2 different locations, that can be utilized as fundamental information for decision making for three levels of strategic management (long term planning), tactic (service adjustment and network development), and operational (scheduling, passenger statistic, and performance indicator). To build O-D matrix is required a predictive model that can be measured to predict passenger destination. The predictive model will be build using classification algorithms such as Decision Tree and K-Nearest Neighbor (KNN).

Sign in / Sign up

Export Citation Format

Share Document