This project is about my Scientific Initiation Research in FEI guided by Wachs-Lopes
Many Computer Vision based applications work with low-level information extracted from natural images. These information can be related to color, texture, shape, spatial position of regions, contrast, distribution of information, among many others. However, in some systems such as object recognition, sometimes considering all these features simultaneously can worsen the hit rate and its performance. Thus, there is a need to choose a set of discriminant features that make the system more precise and reliable. In this paper we propose a bipartite graph model and a metric for the analysis of discriminant features for object recognition systems. Results show that on average, our metric estimates with 95% correlation the performance of an object recognition system.
This Project gave us an article published in Proceedings of WVC 2015 (Computer Vision Workshop) and It was presented in the Oral Session.