Seite - 119 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Fig. 4. Multidimensional scaling layout of all frontal views for parameter- type SmoothnessTerm on Model 0098 using Peak Signal to Noise Ratio (PSNR) as similarity measure and Hue, Saturation, Value colour space. The farther the images are away from each other the more they are different in the meaning of PSNR. Therefore the failed reconstruction on bottom right has a high distance to the other images. A further eventual loss of generality is the coupling to a specific colour space (HSV) together with the assumption that human decisions are dependent on skin deviations. All models in the dataset are Central Europeans with white skincolour. It is not sure that the proposed approach in this configuration works also with models having other skincolours. Nevertheless the approach is a good starting point for future work, see Section VI. A final point regarding applicability is that the proposed approach inspects all parameter-types isolated, see Sec- tion VI on future work to this issue. B. Runtime Considerations The proposed method utilizes a brute force evaluation of all parameter values. While the final comparison of the views is computationally cheap, the reconstruction itself is time consuming: On an Intel Core i5-200 CPU with a NVIDIA Geforece GTX 560 and 16GB RAM it takes in the mean 145s to do a reconstruction. To overcome this issue, the reconstruction has been implemented as web service in the Amazon Cloud. Since the reconstructions are independent of each other, they could be run fully in parallel, benefiting from the virtually infinite computational power in the cloud. However in practice we run the parallelization in a way such that one parameter-type can be fully evaluated at once. VI. CONCLUSION AND FUTURE WORK In this work an approach utilizing pairwise comparison of 2D views from different 3D model reconstructions has been demonstrated, which simulates human quality choices. The approach shows perfect prediction on the given dataset. The essential part of the approach is to select the recon- struction which is most similar to all others. The effect is that reconstructions with local deviations are sorted out. This idea is new and might inspire other scientific work. From the technical side there are two main possibilities of improvement, which are caused by the nature of the used dataset. First the dataset only covers white-skinned Central Europeans and the approach is coupled to a specific colour space. So there could be a loss in generality when inspecting models with other skincolours. To overcome this a future work could use a face detector as pre-step and parametrize the comparison to the actual skincolour. For this new models have to be added to the dataset. Another future work may approach the issue of isolated parameter-type evaluation. Unfortunately with the available questionnaire, combinations of parameters cannot be evalu- ated since they are not in the data. However for future work this would be very interesting, since it could provide further insights to the generality of the approach. In case that there will be significant dependencies between parameter-types a future version may include some kind of genetic algorithm to find the best combination. ACKNOWLEDGMENT This research is carried out within the ”FTI-Project Pro- TechLab” project funded by the State of Upper Austria through the Strategic Economic and Research Program ”In- novatives OO¨ 2020”. REFERENCES [1] C. Heindl, S.C. Akkaladevi, and H. Bauer. Capturing Photorealistic andPrintable 3DModelsUsingLow-CostHardware, pages 507–518. Springer International Publishing, Cham, 2016. [2] Z. Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity. Image Processing, IEEE Transactions on, 13(4):600–612, April 2004. [3] Y. Pan, I. Cheng, and A. Basu. Quality metric for approximating subjectiveevaluationof3Dobjects. IEEETransactionsonMultimedia, 7(2):269–279, April 2005. [4] R.Mekuria,P.Cesar, I.Doumanis, andA.Frisiello. Objectiveandsub- jective quality assessment of geometry compression of reconstructed 3D humans in a 3D virtual room, 2015. [5] J. You, G. Jiang, L. Xing, and A. Perkis.Quality ofVisualExperience for 3D Presentation - Stereoscopic Image, pages 51–77. Springer Berlin Heidelberg, Berlin, Heidelberg, 2010. [6] X. Wang, Q. Liu, R. Wang, and Z. Chen. Natural image statistics based 3D reduced reference image quality assessment in contourlet domain. Neurocomputing, 151, Part 2:683 – 691, 2015. [7] A.RehmanandZ.Wang. Reduced-reference imagequalityassessment by structural similarity estimation. IEEE Transactions on Image Processing, 21(8):3378–3389, Aug 2012. [8] D.S. Alexiadis, A. Chatzitofis, N. Zioulis, O. Zoidi, G. Louizis, D. Zarpalas, and P. Daras. An integrated platform for live 3D human reconstruction and motion capturing. IEEE Transactions on Circuits and Systems for Video Technology, PP(99):1–1, 2016. [9] M. Sedlacek. Evaluation of RGB and HSV models in human faces de- tection. Central European seminar on computer graphics, Budmerice. In IIIA.1-5 - Conference on Computer Systems and Technologies - CompSysTech2004, page 125131, 2004. [10] K.B. Shaik, P. Ganesan, V. Kalist, B.S. Sathish, and J.M.M. Jenitha. Comparative study of skin color detection and segmentation in HSV and YCbCr color space. Procedia Computer Science, 57:41 – 48, 2015. 3rd International Conference on Recent Trends in Computing 2015 (ICRTC-2015). [11] I. Borg and P.J.F. Groenen.ModernMultidimensional Scaling:Theory andApplications (Springer Series in Statistics). Springer, 2nd edition, August 2005. 119