Color histogram assessment of barley under changing daylight conditions

Analysis and interpretation of outdoor images is a difficult problem becaus e of the changing light conditions. If the actual light condition for each individual image could be derived from the image data, the segmentation met hods might be controlled to provide robust results. This paper describes a pilot study on how illumination characteristics may be derived directly fro m image data of Barley using the dichromatic reflection model. Analysis of the colour histogram allows identification of light source thro ugh the observed surface and body reflection from vegetation. Knowing camer a characteristics and using 36 pre segmented images 86% of them were classi fied correctly into four classes, i) sunlight unclouded, ii) sunlight cloud ed, iii) skylight unclouded, and iv) skylight clouded. All "misclassified" images are at the transitions between two classes. The results indicate tha t adaptive image analysis may be supported be the proposed approach.