Evaluation of single and multi-thresholdentropy-based algorithms for folded substrate analysis

Abstract

This paper presents a detailed evaluation of two variants of Maximum Entropy image segmentation algorithm (single and multi-thresholding) with respect to their performance on segmenting test images showing folded substrates. The segmentation quality was determined by evaluating values of four different measures: misclassification error, modified Hausdorff distance, relative foreground area error and positive-negative false detection ratio. New normalization methods were proposed in order to combine all parameters into a unique algorithm evaluation rating. The segmentation algorithms were tested on images obtained by three different digitalisation methods covering four different surface textures. In addition, the methods were also tested on three images presenting a perfect fold. The obtained results showed that Multi-Maximum Entropy algorithm is better suited for the analysis of images showing folded substrates.