Advanced search
Start date
Betweenand

Deep learning and complex networks applied to computer vision

Grant number: 16/18809-9
Support type:Research Grants - Research Partnership for Technological Innovation - PITE
Duration: September 01, 2017 - August 31, 2019
Field of knowledge:Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
Cooperation agreement: IBM Brasil
Principal Investigator:Odemir Martinez Bruno
Grantee:Odemir Martinez Bruno
Home Institution: Instituto de Física de São Carlos (IFSC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Company: IBM Brasil - Indústria, Máquinas e Serviços Ltda
City: São Carlos
Assoc. researchers:Alexandre Souto Martinez ; Joao Batista Florindo

Abstract

Networks have been successfully used in many areas of knowledge that covers practically all fields of Science. The main reason behind the growing interest in networks lies in the fact that it shows a different perspective of the traditional data analysis. During centuries, the scientific research paradigm was ruled by the reductionist approach. Scientific and technological advances increased the amount of data and also encouraged the development of powerful computers, which are capable of processing and storing this huge amount of data. This scenario, often called ''big data'', requires the development of an integrative paradigm of science, that is naturally addressed by networks. During the last decades, Pattern Recognition (PR) has been widely used in both fundamental and applied sciences. Remarkably, most of the PR applications deals with a big amount of data which are difficult handle with the reductionist approach. The combination of PR and networks arises as an important alternative in the big-data scenario for finding, identifying, analyzing, and clustering patterns that are unfeasible to deal with other approaches.Pattern recognition in networks aims at the characterization of networks by extracting information regarding the correlation between vertices and their relationship with topology. This information may lead to the comprehension of network patterns that are intrinsically related to the network model. Due to the very good performance of classification and identifications task, Deep learning is a field that has been attracted the attention of big data and pattern recognition researchers. Although, deep learning methods has been successfully applied in computer vision and signal processing, the combination of deep learning and networks is something very recent, with very few papers into the literature. Nevertheless, it is an interesting and promising research. The proposal of this project is combining the networks and deep learning for pattern recognition purpose in computer vision. In the last decade, the proponent group has been developing methods for computer vision in several problems and obtained very good results with the methods based on complex networks. This way, we choose the computer vision as a case of study of the proposed approach: the combination of deep learning and networks. In computer vision, three specific problems will be investigated: static texture, dynamic texture and boundary shape analysis. The results obtained, can be used immediately to computer vision and has potential to be extended to any application that model the data as networks and demand a good pattern recognition approach. (AU)

Scientific publications (10)
(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)
RIBAS, LUCAS C.; SA JUNIOR, JARBAS JOACI DE MESQUITA; SCABINI, LEONARDO F. S.; BRUNO, ODEMIR M. Fusion of complex networks and randomized neural networks for texture analysis. PATTERN RECOGNITION, v. 103, JUL 2020. Web of Science Citations: 0.
RIBAS, LUCAS C.; MACHICAO, JEANETH; BRUNO, ODEMIR M. Life-Like Network Automata descriptor based on binary patterns for network classification. INFORMATION SCIENCES, v. 515, p. 156-168, APR 2020. Web of Science Citations: 0.
SCABINI, LEONARDO F. S.; RIBAS, LUCAS C.; BRUNO, ODEMIR M. Spatio-spectral networks for color-texture analysis. INFORMATION SCIENCES, v. 515, p. 64-79, APR 2020. Web of Science Citations: 0.
RIBAS, LUCAS C.; BRUNO, ODEMIR M. Dynamic texture analysis using networks generated by deterministic partially self-avoiding walks. PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS, v. 541, MAR 1 2020. Web of Science Citations: 0.
DE MESQUITA SA JUNIOR, JARBAS JOACI; RIBAS, LUCAS CORREIA; BRUNO, ODEMIR MARTINEZ. Randomized neural network based signature for dynamic texture classification. EXPERT SYSTEMS WITH APPLICATIONS, v. 135, p. 194-200, NOV 30 2019. Web of Science Citations: 1.
MESQUITA SA JUNIOR, JARBAS JOACIDE; BACKES, ANDRE RICARDO; BRUNO, ODEMIR MARTINEZ. Randomized neural network based signature for color texture classification. MULTIDIMENSIONAL SYSTEMS AND SIGNAL PROCESSING, v. 30, n. 3, p. 1171-1186, JUL 2019. Web of Science Citations: 0.
SCABINI, LEONARDO F. S.; CONDORI, RAYNER H. M.; GONCALVES, WESLEY N.; BRUNO, ODEMIR M. Multilayer complex network descriptors for color-texture characterization. INFORMATION SCIENCES, v. 491, p. 30-47, JUL 2019. Web of Science Citations: 2.
RIBAS, LUCAS CORREIA; GONCALVES, DIOGO NUNES; SILVA, JONATHAN DE ANDRADE; DE CASTRO, JR., AMAURY ANTONIO; BRUNO, ODEMIR MARTINEZ; GONCALVES, WESLEY NUNES. Fractal dimension of bag-of-visual words. PATTERN ANALYSIS AND APPLICATIONS, v. 22, n. 1, p. 89-98, FEB 2019. Web of Science Citations: 1.
MIRANDA, GISELE H. B.; MACHICAO, JEANETH; BRUNO, ODEMIR M. An optimized shape descriptor based on structural properties of networks. DIGITAL SIGNAL PROCESSING, v. 82, p. 216-229, NOV 2018. Web of Science Citations: 0.
DE MESQUITA SA JUNIOR, JARBAS JOACI; BACKES, ANDRE RICARDO; BRUNO, ODEMIR MARTINEZ. Randomized neural network based descriptors for shape classification. Neurocomputing, v. 312, p. 201-209, OCT 27 2018. Web of Science Citations: 3.

Please report errors in scientific publications list by writing to: cdi@fapesp.br.