| Grant number: | 13/21251-1 |
| Support Opportunities: | Scholarships in Brazil - Master |
| Start date: | April 01, 2014 |
| End date: | February 28, 2015 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science |
| Principal Investigator: | Anderson de Rezende Rocha |
| Grantee: | Alberto Arruda de Oliveira |
| Host Institution: | Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil |
Abstract Among the different types of digital documents, images are of special interest as they carry on a largeamount of semantic information. In fact, the problem of identifying near duplicates of images becamevery popular, with many solutions being proposed and achieving good results in the last decade. Morerecently, some research groups have approached the near duplicate detection problem under a new vantagepoint: the one of tracing back the relations among near duplicates, representing them as a parent-childrelationship. This approach results in a visual structure showing the inheritance of semantic contentamong images giving rise to what is now called image phylogeny. In this project, we aim at extendingupon the original formulation of the image phylogeny problem to deal with situations whereby an imagemay inherit semantic content not only from one parent, but from multiple different parents. This scenariois very common nowadays, in the form of image mosaics, compositions, montages and some other formsthat combine content from different images. Our main objective is to propose a method that is robustenough to deal with the particularities of each case in this context, leading us to a deeper understanding of the relations between not only near duplicates, but also images with complete different content. Identifying multiple parenting relationships in Image Phylogeny presents many challenges, such as correctly finding groups of near duplicates in sets of images with no categorization, localizing shared content between images and comparing the shared content found. With those challenges in mind, we propose to explore three methods: a Forest-, a Clustering- and a Local Dissimilarity- based one. We also aim at developing quality metrics and collecting proper datasets for benchmarking the methods. Finally, this research project will add to the current efforts of our research group in a larger project partially funded by the European Union entitled Reverse Engineering of Audio-Video Content Data (Rewind) formed by researchers from Brazil and European Union. | |
| News published in Agência FAPESP Newsletter about the scholarship: | |
| More itemsLess items | |
| TITULO | |
| Articles published in other media outlets ( ): | |
| More itemsLess items | |
| VEICULO: TITULO (DATA) | |
| VEICULO: TITULO (DATA) | |