Colour Correction for Assessing Plant Pathology using Low Quality Cameras

We describe a framework for standardising the colour of plant images taken using both mobile phones and compact cameras. This is with a view to maximising the accuracy of plant pathology diagnosis. Rather than attempt to characterise each camera, we place cheap and easily portable custom colour targets in the scene being captured. We propose a novel weighted least squares formulation for optimising the transformation function deduced for each image, where the relative contribution for each patch is proportional to the number of closely matching pixels in the image to be transformed. We also introduce our custom colour target which has been designed to preferentially map plant colours and facilitate simple automatic extraction from the image being processed. We provide subjective and objective results demonstrating the efficacy of our approach. It is anticipated that the methods described here could be applied to any application where perceptual consistency