Affordable Access

Access to the full text

Impurities detection in edible bird’s nest using optical segmentation and image fusion

Authors
  • Yee, Cong Kai1
  • Yeo, Ying Heng1
  • Cheng, Lai Hoong2
  • Yen, Kin Sam1
  • 1 Universiti Sains Malaysia, Engineering Campus, Nibong Tebal, Penang, 14300, Malaysia , Nibong Tebal (Malaysia)
  • 2 Universiti Sains Malaysia, Gelugor, Penang, 11800, Malaysia , Gelugor (Malaysia)
Type
Published Article
Journal
Machine Vision and Applications
Publisher
Springer-Verlag
Publication Date
Sep 16, 2020
Volume
31
Issue
7-8
Identifiers
DOI: 10.1007/s00138-020-01124-y
Source
Springer Nature
Keywords
License
Yellow

Abstract

The cleanliness of edible bird’s nest (EBN) is among the determinative factors for market acceptance. As it is meant for human consumption, EBN should be free of any impurities or matter which are foreign to it, such as bird feathers, egg fragments and droppings. However, natural variations in composition, density and thickness impose inconsistency to the level of translucency and colour of EBN, resulting in intensity inhomogeneity in EBN images that substantially reduce the accuracy of the segmentation and detection of impurities. Consequently, the segmentation and detection of impurities, which are essential to visual automation in the cleaning and inspection processes, remain unsolved. This work proposes a novel optical segmentation method to segment impurities from the EBN, in order to facilitate the detection of impurities. EBN images captured under two different lighting scenarios, namely, low-angle blue-diffused lighting and red-diffused backlighting, were used to prepare the fused image for background-EBN-impurities segmentation. The applicability of the method was demonstrated by comparing the detection results with those of human inspectors. With a simple thresholding operation performed on fused images, the impurities detection algorithm recorded a true positive/recall rate of 93.39%, a precision of 71.17% and a false-negative detection rate of 4.8%. Despite the high misclassification rate of 32.25%, the algorithm was able to detect more than 93% of the impurities, compared to human inspectors, who required a second examination on the EBNs.

Report this publication

Statistics

Seen <100 times