Brain-inspired computer vision with applications to pattern recognition and computer-aided diagnosis of glaucoma

Guo, J., 2017, [Groningen]: University of Groningen. 137 p.

Research output: ThesisThesis fully internal (DIV)Academic

Copy link to clipboard


This thesis focuses on developing novel algorithms for visual pattern recognition that are inspired by the shape-selective neurons in the visual cortex. The thesis addresses three applications in computer vision as well as one practical application in ophthalmology.

In the first part of the thesis, we investigate the inhibition phenomenon that occurs in different layers of the visual cortex and then seek to construct computational models for shape detection and object recognition. We then demonstrated the effectiveness of the proposed inhibition-augmented COSFIRE model of shape-selective neurons on three applications, which are the exclusive detection of vascular birurcations in retinal fundus images, the recognition and localization of architectural and electrical symbols, and the recognition of handwritten digits.

In the second part, we focus on developing an automated system that helps ophthalmologists on the population-based glaucoma screening in retinal fundus images. The system consists of localization and delineation of the optic disc, segmentation of the cup and the computation of the vertical cup-to-disc ratio. We evaluate the performance of the proposed approach on eight publicly available data sets and demonstrate its effectiveness and the generalization ability. The system could be deployed as part of a population-based glaucoma screening.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Petkov, Nicolai, Supervisor
  • Azzopardi, George, Co-supervisor
  • Jiang, Xiaoyi, Assessment committee, External person
  • Mandic, D. P. , Assessment committee, External person
  • Telea, Alexandru, Assessment committee
Award date4-Dec-2017
Place of Publication[Groningen]
Print ISBNs978-94-034-0274-1
Electronic ISBNs978-94-034-0273-4
Publication statusPublished - 2017

Download statistics

No data available

ID: 50535293