Deep Image Segmentation Using a Morphological Edge Operator

Abstract

Background: Segmentation of deep images is a difficult, persistent problem in the computer vision field. This paper aimed to address the defects of traditional segmentation methods with deep images, presenting a deep image segmentation algorithm based on a morphological edge operator. Methods: Deep image edge features were first extracted using three traditional edge operators; the edge and tip type jump edges were then extracted via a morphological edge operator, which was used to make the boundary connection; finally, to obtain more accurate segmentation results, skeletonizing was used to refine the image. Results: Compared with traditional segmentation algorithms, the improved algorithm obtained smooth and continuous boundaries, protected edge information from blurring, and was slightly more efficient. When Mickey Mouse depth images were used as experimental subjects, the computational time was reduced by 12.62 seconds; when rabbit depth images were used, computational time was reduced by 17.53 seconds. Conclusion: Morphological edge operator algorithm proposed in this paper is much more effective than traditional edge detection operators algorithms for deep image segmentation; it can clearly divide Mickey Mouse's ears, eyes, pupils, nose, and mouth.