Brain-inspired computing for machine vision

Brain-inspired computing for machine vision

Place: Large Lecture Room

Affiliation: University of Groningen, The Netherlands  

Background: Insights into the function of the visual system of the brain can provide clues for solving computer vision tasks. For instance, the popular Gabor filter was inspired by the function of orientation-selective neurons in areas V1 and V2 of visual cortex. Information about the function of further cortical areas, such as V4 and TEO, has not been sufficiently used yet in computer vision.

Methods: We propose a novel keypointdetector that is inspired by the properties of shape-selective neurons in area V4 of visual cortex. It is trainable as it is configured by the automatic analysis of a feature specified by a user. We configure a set of detectors that are selective for vascular bifurcations in retinal fundus images and demonstrate how such filters can be used to detect similar features. The automatic configuration of such an o nnels of a bank of Gabor filters and the response of the proposed filter is computed as the product of their responses at specific locations. The proposed operator can be implemented using convolutions, shifting, blurring and pixel-wise function evaluation.

Results: With the proposed operators, we achieve a recall rate of 98.52% and a precision rate of 95.19% on a set of 40 binary fundus images from the DRIVE data set containing more than 5000 bifurcations and cross-overs. The SIFT approach achieves a recall rate of 82.04% and a precision rate of 51.87%. We applied the proposed operators to the recognit and achieved results (of 99.45% correct recognition) near the best ever achieved results on the complete MNIST dataset.