Combination of local and global features in image operator learning

• Grant number: 15/17741-9
• Support Opportunities: Regular Research Grants
• Start date: December 01, 2015
• End date: February 28, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Nina Sumiko Tomita Hirata
• Grantee: Nina Sumiko Tomita Hirata
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• City of the host institution: São Paulo
• Associated researchers: Roberto Hirata Junior

Abstract

The problem of designing morphological operators can be modeled in the context of machine learning as a problem of learning a local function that maps the pattern observed in each point of the image to an output value. An interesting characteristic of morphological operators is the fact that they allow an intuitive interpretation of their effects since their conception is strongly based on exploring shape and topological information. Moreover, they are formally well characterized by a sound theoretical foundation. However, by construction, morphological operators do not possess interesting properties such as scale or rotation invariance and they also do not take global or context information into consideration. In this project, the main aim is to advance existing morphological operator learning methods in order to make them able to process objects of different scales and also to take global and context information into consideration. To that end, the main idea to be explored is the use of a variety of feature descriptors cited in literature, coupled to the operator combination framework. Associated theoretical, statistical and practical aspects will be studied. Applications in document image processing are planned as a means to validate the methods to be developed. (AU)

Scientific publications

(7)

(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)

HIRATA, NINA S. T.; MONTAGNER, IGOR S.; HIRATA, ROBERTO, JR.; ACM. Comics image processing: learning to segment text. PROCEEDINGS OF THE 1ST INTERNATIONAL WORKSHOP ON COMICS ANALYSIS, PROCESSING AND UNDERSTANDING (MANPU 2016), v. N/A, p. 6-pg., 2016-01-01. (11/23310-0, 15/01587-0, 15/17741-9, 14/21692-0)

JULCA-AGUILAR, FRANK D.; MAIA, ANA L. L. M.; HIRATA, NINA S. T.; IEEE. Text/non-text classification of connected components in document images. 2017 30TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), v. N/A, p. 6-pg., 2017-01-01. (15/17741-9)

MAIA, ANA L. L. M.; JULCA-AGUILAR, FRANK D.; HIRATA, NINA S. T.; IEEE. A Machine Learning approach for Graph-based Page Segmentation. PROCEEDINGS 2018 31ST SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), v. N/A, p. 8-pg., 2018-01-01. (15/17741-9)

MONTAGNER, IGOR S.; HIRATA, NINA S. T.; HIRATA, ROBERTO, JR.; CANU, STEPHANE; IEEE. Kernel approximations for W-operator learning. 2016 29TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), v. N/A, p. 8-pg., 2016-01-01. (11/50761-2, 15/01587-0, 11/23310-0, 14/21692-0, 15/17741-9)

JULCA-AGUILAR, FRANK D.; HIRATA, NINA S. T.; IEEE. Image operator learning coupled with CNN classification and its application to staff line removal. 2017 14TH IAPR INTERNATIONAL CONFERENCE ON DOCUMENT ANALYSIS AND RECOGNITION (ICDAR), VOL 1, v. N/A, p. 6-pg., 2017-01-01. (15/17741-9)

JULCA-AGUILAR, FRANK D.; HIRATA, NINA S. T.; IEEE. Symbol detection in online handwritten graphics using Faster R-CNN. 2018 13TH IAPR INTERNATIONAL WORKSHOP ON DOCUMENT ANALYSIS SYSTEMS (DAS), v. N/A, p. 6-pg., 2018-01-01. (15/17741-9)

MONTAGNER, IGOR S.; HIRATA, NINA S. T.; HIRATA, JR., ROBERTO. Staff removal using image operator learning. PATTERN RECOGNITION, v. 63, p. 310-320, MAR 2017. (11/23310-0, 15/17741-9, 11/00325-1, 14/21692-0)