<![CDATA[This study aims to develop a blanket pixel-based approach to construct object geometry for image segmentation. Object geometry can be formed from a collection of pixels generated from the edge detection process. However, edge pixels that will be included in a segment must go through an identification process to determine their identity, with a reference segment as a reference for labeling. This work proposes the terminology of blanket pixels, namely pixels that surround a pixel that does not yet have an identity due to being isolated from the surrounding segments, as a spatial exoskeleton for the labeling process. This approach has been tested, and the results show that we successfully detect the structure of tilapia egg circles with clear fortifications when the scanning radius parameter is set to 10 pixels and the proximity between the surrounding pixels and the labeled pixels is 11.8 pixels. Out of 114 egg circles, this method successfully detected 105 eggs, with 9 small eggs (2–3 pixels in diameter) undetected, resulting in a detection ratio of 92.11%. The blanket pixel approach effectively recognizes and reclassifies isolated pixel labels. This approach supports the process of labeling pixels in areas with significant ambiguity.]]>
Copyrights © 2025
