Page - 181 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Dataset Type Vertices Runtime Chars4Testing measured 10,492 0.044s cuneusideal synthetic 15,521 0.071s Half4Testing measured 282,428 0.789s HOSG10Preview measured 371,711 1.809s VAT 10908 measured 3,034,899 12.269s HOSG10Full measured 6,596,964 33.319s Table 1: Performanceof the algorithmonmultiple data sets. Dataset name, type, number of vertices and the respective runtimearegiven. Figure 3: Detected ridge points in the realworld data setChars4Testing. It can be seen that the points followthe ridgesof themeshnicely. (a) (b) Figure4: High resolutiondatasetHOSG10Full (a)beforeand (b)after smoothing. 4. OutlookandSummary Futureenhancementsofouralgorithmaretheimplementationofamarchingfront toconnect theridge points to lines,makingthemexportableasSVG.FollowingtheCannyapproach,hysteresis trackingis a futureextensionprovidinganevenmore robust selectionof featurepoints. Furthermore smoothing of the function values instead of smoothing themeshwill improve the performance by reducing the computational overhead of processingM2. The final vision is to have a completely autonomous system, which begins transcribing the ancient tablets immediately after their acquisition, exporting thedigital drawingswithautomatedannotationsdirectly intoa searchabledatabase [6]. The algorithm implemented in this work succeeds in extracting ridge points from irregular grids, using non-maximumsuppression. Although the execution on an irregular surfacemesh architecture 181