Multi-object sketch segmentation using convolutional object detectors

Momina Moetesum; Osama Zeeshan; Imran Siddiqi

doi:10.1117/12.2524293

6 May 2019 Multi-object sketch segmentation using convolutional object detectors

Momina Moetesum, Osama Zeeshan, Imran Siddiqi

Proceedings Volume 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018); 1106929 (2019) https://doi.org/10.1117/12.2524293
Event: Tenth International Conference on Graphic and Image Processing (ICGIP 2018), 2018, Chengdu, China

Abstract

Segmentation of constituent shapes is a vital yet challenging step in any automated sketch analysis and interpretation system. Conventional stroke level sketch segmentation techniques perform well on a single shape, nevertheless, their performance degrades in a cluttered multi-object sample. On the contrary, object level techniques rely on proximity based perceptual grouping for shapes with disjoint constituent parts, which requires high level semantic knowledge. To overcome these challenges, we propose the use of state-of-the-art convolutional object detectors for the detection and segmentation of hand drawn shapes from offline samples of a neuropsychological drawing test i.e. Bender Gestalt Test (BGT). Experiments with different combinations of convolutional meta-architectures and feature extractors show that such networks can successfully be employed for sketch segmentation purposes even with limited training data and resources. Amongst all network combinations under evaluation in this study, Faster R-CNNs with ResNet-101 as feature extractor, outperform others by achieving precision, recall and F-measure values of 92.93%, 95.24% and 94.07% respectively.

Citation Download Citation

Momina Moetesum, Osama Zeeshan, and Imran Siddiqi "Multi-object sketch segmentation using convolutional object detectors", Proc. SPIE 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018), 1106929 (6 May 2019); https://doi.org/10.1117/12.2524293

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available