Minimum Error Thresholding Research Articles

On Threshold Selection Using Fuzzy Risk Criterion

A parametric and unsupervised method of automatic threshold selection for image segmentation is presented. An optimal threshold is selected by the fuzzy risk criterion, namely, so as to separate a given picture into meaningful gray level classes under the assumption of object and pixel gray level values being normally distributed. The effectiveness of the proposed approach is illustrated using some experimental images with 256 gray-levels, and the results are compared with minimum error thresholding method using the criterion functions such as uniformity, shape, and edge measures.

A note on minimum error thresholding

In minimum error thresholding originated by Kittler and Illingworth, we present a new expression of the minimum error criterion by the Shannon entropy.

Improvement of kittler and illingworth's minimum error thresholding

A simple modification to Kittler and Illingworth's minimum error thresholding method was made and the performance of the modified version was compared with that of the original version empirically. By correcting the biased estimates of variances of model distributions, a significant improvement in performance was found. The improvement was most outstanding among not-well-separated, but still bimodal histograms. In fact, the modification provides a more robust method. The new version is nearly computationally equivalent in complexity to the original version.

On minimum error thresholding and its implementations

In this paper we discuss the thresholding methods [1] due to Kittler and Illingworth. Two methods are suggested in [1], namely, 1) minimizing a criterion function aand 2) iterative search of the minimum error threshold. We have tested the two methods. The first one is found to be robust even in the case of weak bimodality in the histogram. The second one is faster than the first, but a good initial threshold must be provided. We have proposed to use the method of Ridler [3] and Trussel [4] to select the initial threshold. Experimental results show that if the initial threshold is selected this way the second method is as robust as the first one, but the speed of the whole procedure is much faster.

Minimum error thresholding

A computationally efficient solution to the problem of minimum error thresholding is derived under the assumption of object and pixel grey level values being normally distributed. The method is applicable in multithreshold selection.