• home
• current students
• graduate
• events

Presenter: Alireza Rezvanifar
Supervisor:

Date: Thu, December 20, 2018
Time: 10:30:00 - 11:30:00
Place: EOW 430

ABSTRACT

Abstract:

The analysis of digital graphics-rich document images, such as architectural floor plans, focuses on the automatic extraction of visual information which are initially addressed to human comprehension. Symbol spotting, a rather newly emerged term in this field, is defined as finding a ranked list of regions of interest (ROIs) in an input image which are more likely to carry a query symbol without explicitly recognizing that symbol.

In this presentation, we first present a review of the recent progress on symbol spotting with a specific focus on digital architectural floor plans as an application. An architectural floor plan (or architectural drawing) is a scaled two-dimensional diagram of one level of a building: consisting of lines, symbols, and textual markings. Next, a new geometric descriptor based on an ellipse detector is proposed to capture the most important geometric shapes constituting symbols. Finally, a spotting approach is introduced to integrate both statistical and structural approaches for finding symbols in the image. This way, not only can we take the advantages of structural approaches for representing spatial information of symbols, but also the statistical method which will help us to reduce computational complexity of the spotting step.