Reclassification of segmentation boundary base on neighboring function

Motivated by the goal of improving the performance of segmentation, a new technique based on Neighboring Function (NF) is presented to reclassify the rough segmentation boundary pixels. The NF is a novel measurement of neighboring relationship, it takes into consideration of both spatial and intensity information and their distribution pattern. With the rough boundary provided by other segmentation algorithms, the value of the NF at each boundary pixel is calculated in a specific neighborhood, then the reclassification will be implemented by comparing these values. In order to obtain the expected boundary, this step is iterated for several times. In our study, the proposed method is applied to synthetic and real medical images, a great improvement of the quality of segmentation boundary has been achieved. The accuracy and reproducibility of this reclassification method has been proven by experimental results. Experiments also show that this method is insensitive to noise.