A new approach to the determination of unbiased thresholds for image segmentation

This paper describes a new approach to the determination of intensity thresholds for image segmentation: it is based on interpolating between two or more sub-optimal thresholds to obtain a single unbiased value. We have found that the remanent bias following interpolation is virtually undetectable over a wide range of peak heights. The particular advantage of the approach is that it avoids the high computation involved in rigorous modelling of intensity distributions by curve fitting, which can prove prohibitive when adaptive thresholding has to be done in real time.<
>