Improving plant discrimination in image processing by use of different colour space transformations

Different transformations of RGB colour space (discriminant analysis, canonical transformation, i1i2i3, HSI, HSV and Lab) were compared to find the best method for separating plants and background in colour images taken by a digital photo camera. Logarithmic discriminant analysis, with misclassification of plant and background pixels of about 2%, proved to be the most suitable transformation. However, for real-time working it will be necessary to develop a sufficiently powerful computer system. In the meantime, the i1i2i3 colour space or the optimised colour space (i1i2i3(new)) can be used in combination with an automatic binarisation system. (C) 2002 Elsevier Science B.V. All rights reserved.