Abstract: Image segmentation is a fundamental problem in machine vision.Image segmentation algorithm based on threshold depends on the parameter adjustment,which is vulnerable to local minimum value and needs a lot of time.It reduces the quality and efficiency of segmentation algorithm.In order to realize the adaptive threshold selection in the process of image segmentation,a novel image segmentation algorithm via adaptive threshold selection and sparse principal component analysis was proposed.According to the content of the image,the algorithm removes the noise with the image noise level obtained by the sparse principal component analysis.The global segmentation threshold is obtained by the main region of the image based on 2D histogram.Then the local segmentation threshold is obtained by local details of image based on moving average method.Finally,the global threshold segmentation and the local threshold segmentation image are combined to obtain the best segmentation results.The simulation and experimental results on Berkeley data set show that the algorithm has an advantage on the accuracy edge of image segmentation and robustness to noise compared to current frontier algorithm.It has better segmentation performance on subjectivity and objectivity,and improves the quality of image segmentation.

Key words: Threshold segmentation,Sparse principal component analysis,Global threshold,Local threshold