An adaptive multi-threshold segmentation algorithm for complex images under unstable imaging environment

Abstract

Images acquired from actual manufacturing practice are difficult to segment for uncertainty surface quality of work-piece and unstable imaging environment, which may result in limited robustness for traditional image segmentation algorithms. Nevertheless, the result of image segmentation will affect the precision of subsequent feature extraction, image analysis and robot positioning. To address the above issues, an adaptive multi-threshold segmentation algorithm for complex images under unstable imaging environment is proposed to further improve the stability of an industrial vision system in a given inspection scheme. The proposed approach consists of three basic parts and each of them is indispensable for achieving high accuracy. Firstly, curve fitting with cubic spline interpolation to determine peaks and troughs of grey histogram by calculating function extremums, and then the higher greyscale between top-peak and sub-peak is selected as the initial lower threshold. Secondly, the initial upper threshold is constrained by calculating the probability density distribution of ROI. Thirdly, the upper threshold and lower threshold are iteratively calculated until the threshold range of ROI is achieved. The proposed algorithm was compared to state-of-the-art segmentation approaches in both synthetic and real images to demonstrate its superior performance.