Qualitative and Quantitative Evaluation of Two New Histogram Limiting Binarization Algorithms

摘要

Image segmentation and thus feature extraction by binarization is a crucial aspect during image processing. The "most" critical criteria to improve further analysis on binary images is a least- biased comparison of different algorithms to identify the one performing best. Therefore, fast and easy-to-use evaluation methods are needed to compare different automatic intensity segmentation algorithms among each other. This is a difficult task due to variable image contents, different histogram shapes as well as specific user requirements regarding the extracted image features. Here, a new color-coding-based method is presented which facilitates semi-automatic qualitative as well as quantitative assessment of binarization methods relative to an intensity reference point. The proposed method represents a quick and reliable, quantitative measure for relative binarization quality assessment for individual images. Moreover, two new binarization algorithms based on statistical histogram values and initial histogram limitation are presented. This mode-limited mean (MoLiM) as well as the differential-limited mean (DiLiM) algorithms were implemented in ImageJ and compared to 22 existing global as well as local automatic binarization algorithms using the evaluation method described here. Results suggested that MoLiM quantitatively outperformed 11 and DiLiM 8 of the existing algorithms.