Adaptive computation of unimodal and hysteresis thresholds

This paper proposes a new histogram-based dual-threshold computation algorithm for the image of unimodal histogram in order to improve the adaptive acquisition of low-and high-threshold for the unimodal hysteresis thresholding. The novel bowstring is firstly introduced to make an accurate priori-measurement of the overall tendency of the histogram. Based on this global information, the dual-threshold is further computed by adaptively searching two tangent points corresponding to the properly defined transitional characteristics over the whole histogram. The effectiveness of this proposed algorithm is evaluated from two aspects such as the unimodal and hysteresis thresholding using the Baddeley's discrepancy and receiver operating characteristics (ROC) curve evaluation method, respectively. The comparisons to the MSA, Medina-Carnicer's and Otsu's method are presented in detail and demonstrate the feasibility and reliability of the proposed dual-threshold computation approach.