There are provided a system and method of defect detection on a specimen, the method comprising: performing partitioning for each of one or more portions of a first die; receiving one or more noise maps indicative of noise distribution on second images captured for one or more portions of a second die; performing segmentation for each noise map in runtime, the segmentation for a given noise map including: calculating a score for each region, the given noise map aligned with the regions and each region is associated with noise data aligned therein, the score for a given region calculated at least based on the noise data associated therewith; and associating each region with one segmentation label of a predefined set of segmentation labels indicative of noise levels based on the score, thereby obtaining a set of segments each corresponding to one or more regions associated with the same segmentation label.