Effects of Color Characterization on Computational Efficiency of Feature Detection with Live-Object Handling Applications
Qiang Li, Kok-Meng Lee · 2005
This paper presents a machine vision algorithm that utilizes the principal component analysis technique to characterize target features in color space from a set of training data so that the color classification can be done accurately and efficiently. The method, referred to here as the statistically based fast bounded box (SFBB), has significant potential in agriculture and food processing applications where color variability often renders grayscale-based algorithms difficult or impossible to work. We evaluate the algorithm in the context of live-bird handling applications and examine the effects of the color characterization on computational efficiency by comparing the proposed solution against two commonly used color classification algorithms; the RCE neural network classifier and the support vector machine. Comparison among the three methods demonstrates that SFBB is relatively easy to train, efficient and effective since with sufficient training data it requires no additional optimization steps; these advantages make SFBB an ideal candidate for high-speed automation involving live and/or natural objects.