Relief Extraction From a Rough Stele Surface Using SVM-Based Relief Segment Selection

A Machine Learning Method Based on 3D Local Surface Representation for Recognizing the Inscriptions on Ancient Stele

“…As illustrated in Figure 5, the characters inscribed on the surface are very difficult to interpret by visual inspection, especially at the center and the lower right portions of the stele. For further information on the Musul-ojakbi stele, we refer the reader to the previous study [24]. To reduce the presence of the degraded region in the ground truth dataset before extracting important features, we carefully cropped fine partitions of the 3D scanned data using the MeshLab 3D processing tool [34].…”

“…Among the machine learning models with acceptable performance in the extraction of reliefs is the SVM-based approach [24]. The main idea of this method involves the initial extraction of relief candidate segments which includes actual reliefs and relief-like noise (cracks/dents).…”

“…The geometrical information was learned by the SVM to determine whether the candidate is relief or not. While this machine learning method outperformed the height estimation [8] and curvature-based methods [7,18], the superior performance of the SVM-based method [24] is largely dependent on the local surface features that were identified on the 3D mesh and the conditions for the removal of degraded region.…”

“…More recently, support vector machine (SVM) was applied to extract relief region from the surface of stone monument [24]. The modified curvature-based method [18] was initially used to identify relief candidate segments from mesh partitions.…”

“…As such, the proposed method classifies the surface points as relief or non-relief region. In contrast with the SVM-based method which classifies relief using candidate segment [24], the classification of the proposed machine learning-based method is done per vertex.…”