Papers by Jean-Charles Créput
Matlab GUI Application for Moving Object Detection and Tracking
Advances in intelligent systems and computing, 2019
In this paper a novel tool for moving object detection and tracking is presented. The main contri... more In this paper a novel tool for moving object detection and tracking is presented. The main contribution of the proposed application is the achievement of a simple and intuitive graphic interface during the extraction the silhouette of targets by means of a new algorithm. This proposed algorithm which combined frame difference method, background subtraction method, Laplace filter and Canny edge detector together can realize a way to achieve sparse detection fast. Some modular architecture in this Graphical User Interface has been developed in order to enhance the user’s experience. The experiment was tested by using sequence images from the MULTIVITION dataset, and experimental results showed that our proposed method has more validity and flexibility to get the desired result than conventional algorithm.
World Academy of Science, Engineering and Technology, International Journal of Electronics and Communication Engineering, Mar 11, 2017
To accelerate the solution for large scale traveling salesman problems (TSP), a parallel 2-opt lo... more To accelerate the solution for large scale traveling salesman problems (TSP), a parallel 2-opt local search algorithm with simple implementation based on Graphics Processing Unit (GPU) is presented and tested in this paper. The parallel scheme is based on technique of data decomposition by dynamically assigning multiple K processors on the integral tour to treat K edges' 2-opt local optimization simultaneously on independent sub-tours, where K can be user-defined or have a function relationship with input size N. We implement this algorithm with doubly linked list on GPU. The implementation only requires O(N) memory. We compare this parallel 2-opt local optimization against sequential exhaustive 2-opt search along integral tour on TSP instances from TSPLIB with more than 10000 cities.
Massively parallel GPU computing for fast stereo correspondence algorithms
Journal of Systems Architecture, Apr 1, 2016
A GPU-based implementation of the fixed window partially demosaiced CFA stereo matching applicati... more A GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application is presented.Accelerations up to 20 times are obtained for large size image pairs.A GPU-based implementation of the adaptive window color stereo matching application is presented.It can handle four pairs of standard images from Middlebury database within roughly 100 ms.A comparative study among different GPU-based local dense stereo matching implementations is provided. Current accurate stereo matching algorithms employ some key techniques that are not suitable for parallel GPU architecture. It will be tricky and cumbersome to directly take these techniques into GPU applications. Trying to tackle this difficulty, we design two GPU-based stereo matching algorithms, one using a local fixed aggregation window whose size is configurable, and the other using an adaptive aggregation window which only includes necessary pixels. We use the winner-takes-all (WTA) principle for optimization and a plain voting refinement for post-processing; both do not need complex data structures. We aim to implement on GPU platforms fast stereo matching algorithms that produce results with same-level quality as other WTA local dense methods that use window-based cost aggregation. In our GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application, accelerations up to 20 times are obtained for large size images. In our GPU-based implementation of the adaptive window color stereo matching application, experiment results show that it can handle four pairs of standard images from Middlebury database within roughly 100 ms.
Massive 2-opt and 3-opt Moves with High Performance GPU Local Search to Large-Scale Traveling Salesman Problem
Lecture Notes in Computer Science, Dec 31, 2018
2-opt, 3-opt or \(k-\)opt heuristics are classical local search algorithms for traveling salesman... more 2-opt, 3-opt or \(k-\)opt heuristics are classical local search algorithms for traveling salesman problems (TSP) in combinatorial optimization area. This paper introduces a judicious decision making methodology of offloading which part of the \(k-\)opt heuristic works in parallel on Graphics Processing Unit (GPU) while which part remains sequential, called “multiple \(k-\)opt evaluation, multiple \(k-\)opt moves”, in order to simultaneously execute, without interference, massive 2-/3-opt moves that are globally found on the same TSP tour or the same Euclidean space for many edges, as well as keep high performance for GPU massive \(k-\)opt evaluation. We prove the methodology is judicious and valuable because of our originally proposed sequential non-interacted 2-/3-exchange set partition algorithm taking linear time complexity and a new TSP tour representation, array of ordered coordinates-index, in order unveil how to use GPU on-chip shared memory to achieve the same goal as using doubly linked list and array of ordered coordinates for parallel \(k-\)opt implementation. We test this methodology on 22 national TSP instances with up to 71009 cities and with brute initial tour solution. Average maximum 997 non-interacted 2-opt moves are found and executed on the same tour of ch71009.tsp instance in one iteration of our proposed method. Experimental comparisons show that our proposed methodology gets huge acceleration over both classical sequential and a possible current fastest state-of-the-art GPU parallel 2-opt implementation.
Moving Object Detection and Tracking Based on Three-Frame Difference and Background Subtraction with Laplace Filter
Lecture Notes in Computer Science, 2018
Moving object detection and tracking is an important research field. Currently, ones of the core ... more Moving object detection and tracking is an important research field. Currently, ones of the core algorithms used for tracking include frame difference method (FD), background subtraction method (BS), and optical flow method. Here, authors are looking at the first two approaches since very adequate for very fast real-time treatments whereas optical flow has higher computation cost since related to a dense estimation. Combination of FD and BS with filters and edge detectors is a way to achieve sparse detection fast. This paper presents a tracking algorithm based on a new combination of FD and BS, using Canny edge detector and Laplace filter. Laplace filter occupies a leading role to sharpen the outlines and details. Canny edge detector identifies and extracts edge information. Morphology processing is used to eliminate interfering items finally. Experimental results show that 3FDBD-LC method has higher detection accuracy and better noise suppression than current combination methods on sequence images from standard datasets.
IEEE Transactions on Visualization and Computer Graphics, Sep 1, 2015
The goal of structured mesh is to generate a compressed representation of the 3D surface, where n... more The goal of structured mesh is to generate a compressed representation of the 3D surface, where near objects are provided with more details than objects far from the camera, according to the disparity map. The solution is based on the Kohonens Self-Organizing Map algorithm for the benefits of its ability to generate a topological map according to a probability distribution and its potential to be a natural massive parallel algorithm. The disparity map, which stands for a density distribution that reflects the proximity of objects to the camera, is partitioned into an appropriate number of cell units, in such a way that each cell is associated to a processing unit and responsible of a certain area of the plane. The advantage of the proposed model is that it is decentralized and based on data decomposition. The required processing units and memory are with linearly increasing relationship to the problem size. Experimental results show that our GPU implementation is able to provide near real-time performance with small size disparity maps and the running time increases in a linear way with a very weak increasing coefficient. The proposed method is suitable to deal with large scale problems in a massively parallel way.
Massively parallel cellular matrix model for self-organizing map applications
We propose the concept of parallel cellular matrix which partitions the Euclidean plane defined b... more We propose the concept of parallel cellular matrix which partitions the Euclidean plane defined by input data into an appropriate number of uniform cell units. Each cell is responsible of a certain part of the data and the network of the self-organizing map (SOM), and carries out massive parallel spiral searches based on the cellular matrix topology. The advantage of the proposed model is that it is decentralized and based on data decomposition. The required processing units and memory are with linearly increasing relationship to the problem size. Based on the cellular matrix model, the parallel SOM is implemented to deal with various applications including the traveling salesman problem, structured mesh generation, and superpixel adaptive segmentation map. Experimental results of our GPU implementation show that the running time increases in a linear way with a very weak increasing coefficient according to the input size. The proposed cellular matrix model is suitable to deal with large scale problems in a massively parallel way.
Self-organizing maps and full GPU parallel approach to graph matching
Computer Communications, 2023
Stereo Matching by Using Self-distributed Segmentation and Massively Parallel GPU Computing
Lecture Notes in Computer Science, 2016
As an extension of using image segmentation to do stereo matching, firstly, by using self-organiz... more As an extension of using image segmentation to do stereo matching, firstly, by using self-organizing map (som) and K-means algorithms, this paper provides a self-distributed segmentation method that allocates segments according to image’s texture changement where in most cases depth discontinuities appear. Then, for stereo, under the fact that the segmentation of left image is not exactly same with the segmentation of right image, we provide a matching strategy that matches segments of left image to pixels of right image as well as taking advantage of border information from these segments. Also, to help detect occluded regions, an improved aggregation cost that considers neighbor valid segments and their matching characteristics is provided. For post processing, a gradient border based median filter that considers the closest adjacent valid disparity values instead of all pixels’ disparity values within a rectangle window is provided. As we focus on real-time execution, these time-consumming works for segmentation and stereo matching are executed on a massively parallel cellular matrix GPU computing model. Finaly, we provide our visual dense disparity maps before post processing and final evaluation of sparse results after post-processing to allow comparison with several ranking methods top listed on Middlebury.
Using Entropy and Marr Wavelets to Automatic Feature Detection for Image Matching
2019 15th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS)
Image matching, also refereed as feature point matching, is a fundamental issue in computer visio... more Image matching, also refereed as feature point matching, is a fundamental issue in computer vision. In this paper, we propose to use local entropy based on Marr wavelets within scale-interaction to improve the accuracy of automatic feature detection in the context of image matching. The goal is to improve the accuracy of the feature matching step while exhibiting a highly representative set of features of the objects within both images. To improve reliability, we propose to exploit local entropy under a mesh division strategy in combination with a sensitive feature selection stage. Experimental results show that this algorithm can outperform some of the conventional feature extraction algorithms with higher subsequent matching recall rate of image matching.
Matlab GUI Application for Moving Object Detection and Tracking
Advances in Intelligent Systems and Computing, 2019
In this paper a novel tool for moving object detection and tracking is presented. The main contri... more In this paper a novel tool for moving object detection and tracking is presented. The main contribution of the proposed application is the achievement of a simple and intuitive graphic interface during the extraction the silhouette of targets by means of a new algorithm. This proposed algorithm which combined frame difference method, background subtraction method, Laplace filter and Canny edge detector together can realize a way to achieve sparse detection fast. Some modular architecture in this Graphical User Interface has been developed in order to enhance the user’s experience. The experiment was tested by using sequence images from the MULTIVITION dataset, and experimental results showed that our proposed method has more validity and flexibility to get the desired result than conventional algorithm.
Massively parallel GPU computing for fast stereo correspondence algorithms
Journal of Systems Architecture, 2016
A GPU-based implementation of the fixed window partially demosaiced CFA stereo matching applicati... more A GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application is presented.Accelerations up to 20 times are obtained for large size image pairs.A GPU-based implementation of the adaptive window color stereo matching application is presented.It can handle four pairs of standard images from Middlebury database within roughly 100 ms.A comparative study among different GPU-based local dense stereo matching implementations is provided. Current accurate stereo matching algorithms employ some key techniques that are not suitable for parallel GPU architecture. It will be tricky and cumbersome to directly take these techniques into GPU applications. Trying to tackle this difficulty, we design two GPU-based stereo matching algorithms, one using a local fixed aggregation window whose size is configurable, and the other using an adaptive aggregation window which only includes necessary pixels. We use the winner-takes-all (WTA) principle for optimization and a plain voting refinement for post-processing; both do not need complex data structures. We aim to implement on GPU platforms fast stereo matching algorithms that produce results with same-level quality as other WTA local dense methods that use window-based cost aggregation. In our GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application, accelerations up to 20 times are obtained for large size images. In our GPU-based implementation of the adaptive window color stereo matching application, experiment results show that it can handle four pairs of standard images from Middlebury database within roughly 100 ms.
IEEE Transactions on Visualization and Computer Graphics, 2015
The goal of structured mesh is to generate a compressed representation of the 3D surface, where n... more The goal of structured mesh is to generate a compressed representation of the 3D surface, where near objects are provided with more details than objects far from the camera, according to the disparity map. The solution is based on the Kohonens Self-Organizing Map algorithm for the benefits of its ability to generate a topological map according to a probability distribution and its potential to be a natural massive parallel algorithm. The disparity map, which stands for a density distribution that reflects the proximity of objects to the camera, is partitioned into an appropriate number of cell units, in such a way that each cell is associated to a processing unit and responsible of a certain area of the plane. The advantage of the proposed model is that it is decentralized and based on data decomposition. The required processing units and memory are with linearly increasing relationship to the problem size. Experimental results show that our GPU implementation is able to provide near real-time performance with small size disparity maps and the running time increases in a linear way with a very weak increasing coefficient. The proposed method is suitable to deal with large scale problems in a massively parallel way.
A Multi-Agent Approach to Adaptive Mesh Generation
ABSTRACT -
International Journal of Computer & Software Engineering, 2017
We study a combinatorial optimization problem for conflict-free routing in a Network-on-Chip. Bas... more We study a combinatorial optimization problem for conflict-free routing in a Network-on-Chip. Based on time division multiplexing and cyclic emission, the problem consists in finding a set of K shortest paths, such that packets will never conflict through the network but can use shared communication links in an efficient way. The model allows to avoid collisions and deadlocks in irregular network topologies, while minimizing latency. A time-expanded graph approach is retained for the solution process. First, we present a mixed integer linear programming model for the problem. Second, a set of shortest paths operators are combined within three iterated local search schemes able to quickly generate admissible solutions for the problem. To evaluate the method, experiments are conducted on a set of five real-life problem instances, and on many artificial unstructured random instances derived from them. We detail the problem of traffic instance generation, that also illustrates the designer's task of flow decomposition between communicating components. Intensive simulations illustrate the accuracy of the solution method.
Http Www Theses Fr, 2012
Problèmes de plus courts chemins dans les NoC et leurs extensions aux cas difficiles Thèse souten... more Problèmes de plus courts chemins dans les NoC et leurs extensions aux cas difficiles Thèse soutenue le 03 décembre 2012, devant la commission d'examen composée de :
Local search methods for conflict-free routing in a multi-processor system on chip
International audienc
An information retrieval system using a new neural network model
In most present information retrieval systems (IRS) with neural networks, the document neurons ar... more In most present information retrieval systems (IRS) with neural networks, the document neurons are directly connected with indexing term neurons by synaptic weights. Each document neuron computes a linear similarity measure: the scalar product of vector space model. We present a model with a hidden layer of neurons. At the present time, back propagation learning (BPL) algorithm requires a high number of iterations, and it does not give a method to find the required number of neurons in the hidden layer. Our model has a learning algorithm based on core vertices and the majority logic. In the cases of non-linearly separable sets of input data, the learning algorithm creates the required hidden neurons. The learning algorithm is applied to a set of couples of input and output vectors: the input binary data vector (the selected and non-selected terms) and the output binary data vector (the selected document). This set is the valuations of the judgements for the IRS. Then this set is mem...
Generic parallel data structures and algorithms to GPU superpixel image segmentation
Displays
Memory Efficient Parallel 2-Opt for Large Scale Travelling Salesman Problems
World Academy of Science, Engineering and Technology, International Journal of Electronics and Communication Engineering, 2017
Uploads
Papers by Jean-Charles Créput