Relevance image sampling from collection using importance selection on randomized optimum-path trees

Author(s)

Ponti, Moacir A. ; IEEE

• Document type: Journal article
• Source: 2017 6TH BRAZILIAN CONFERENCE ON INTELLIGENT SYSTEMS (BRACIS); v. N/A, p. 6-pg., 2017-01-01.

Abstract

The growth in image collections became an important issue when designing a successful image retrieval and recognition system. While it is important to investigate methods that uses smaller training sets or under samples the data, it is also challenging to be successful with a single model trained with a reduced number of samples, since they often require representative and sufficient observations to be accurate. We propose an algorithm that selects relevant images from a collection, based on pasting of small votes ensembles of optimum-path forest base classifiers. Since small training sets are used, it is viable for large datasets. Also, the classifiers tested maintained in general their performances after sampling using our method, even using significantly less training data. (AU)

• FAPESP's process: 16/16111-4 - Feature learning applied to sketch-based image retrieval and low-altitude remote sensing
• Grantee: Moacir Antonelli Ponti
• Support Opportunities: Regular Research Grants