Efficient circular-shape object segmentation method for adjacent objects

Abstract

The general object recognition method is based on the various area segmentation algorithm. With these methods, segmentation is not very difficult if the boundaries between objects are clear, but if the boundaries are vague, the segmentation of the adjacent objects becomes inaccurate. So in order to solve this problem, we propose an efficient method of dividing adjacent circular-shape objects into single object. For segmentation into single object, the final segmentation object is determined in the following three steps: detection of the ROI, determination of the candidate segmentation points, and creation of a segmentation boundary. As a result, robust performance of average 6.2 % difference compared to the existing methods were derived in the experiments, even with severe SNR case.