Thresholding of images in combination with 'improved MAX-MIN filters'

Abstract

In the framework of the EC-funded AIR-project Objective plant quality measurement by digital image processing taking images each three weeks during more than nine months follows the development of Ficus benjamina plants. From these images a large number of features is extracted and a relation is laid between these features and the external quality by using neural networks. A segmentation procedure for classifying the pixels into object pixel (plant) and non-object pixels (background) has to be used before feature extraction. Segmentation procedures based on thresholding depend on the specific threshold that is used, especially when the transition between object and background follows a ramp instead of a step andlor the intensity of the object andlor background is not constant for the whole image. Improved versions of MAX-MIN filters for edge enhancement are less noise sensitive than other filters for edge enhancement as for example a Laplace operator. The same feature extraction procedures are applied to images with different illumination levels and that have or have not been enhanced by improved MAX-MIN filtering. The influence of image enhancement by improved MAX-MIN-filtering on the segmentation of images and consequently on the feature values will be discussed.