Abstract

The use of outdoor machine vision has become part of the technology used in industry, farming, and military. Applications include color recognition such as obstacle detection, road following, and landmark recognition. This study proposes a spike auto-detection and correction technique based on color modeling and surface reflectance to predict the color and correct the spike region apparent color on the tomato surface. This algorithm classifies tomatoes in red, orange, and green color category based on training images with accuracy of 94%. Then by the use of mean shift color segmentation algorithm, the spiky pixels on the surface of tomato are spotted. Based on the color model and Normalized Photometric Function (NPF) for relevant tomato in a tropical place as Malaysia, the color of each spiky pixel is estimated in HSV (hue, saturation, and value) color space. Finally, the specular effects are corrected through replacing their estimated color. From the experimental results, this study demonstrates overall accuracy of 93%. The contribution of the paper lies in the use of outdoor color based models for tropical places as previously developed by the authors to correct the specular effects on a spherical surface such as tomato.

Key words: Color imaging, colorimetry, color models, glossy reflection, surface reflection functions.