An improved moment-preserving auto threshold image segmentation algorithm

Abstract

When moment-preserving auto threshold algorithm is used to segment image whose histogram is unimodal or monotonic function, there is serious background interference, and the segmentation accuracy is greatly affected by the size variance of the object, so this paper puts forward an improved moment-preserving auto threshold algorithm. Aiming at the original algorithm's shortage of neglecting image details, this algorithm takes advantage of the feature that the grey level difference between object borders and adjacent background is great while the difference among pixels in an object region or background region is small, and then adds gradient adjustment based on object edge pixels to moment-preserving auto thresholding, in order to look after both the whole and the details of image in segmentation result. This algorithm needs no iteration or search, and it is fast enough to satisfy the demand for real time. As simulation results show, this algorithm can segment object image effectively.