Fast quality assessment of 3D printed surfaces based on structural similarity of image regions

Abstract

Automatic visual quality evaluation of 3D printed surfaces is one of the recent challenging tasks related to the applications of image quality assessment methods. One of the most desired solutions is a reliable classification of printed surfaces into high and low quality samples during the printing process as it may allow saving the material (filament), power and time. Assuming the side-view location of the camera observing the printed objects the consecutive layers can be observed and their regularity can be assessed using specialized image analysis algorithms. One of the possible approaches, which is considered in this paper, is the application of widely known image quality assessment method, namely Structural Similarity (SSIM). As the original SSIM is the full-reference metric which requires the use of the reference image which is unavailable in our case, the mutual similarity of image regions has been utilized. To decrease the computational complexity only some of the regions have been compared chosen randomly using the Monte Carlo approach with additional region matching. Obtained results are encouraging and the proper classification has been obtained for almost all samples used in experiments.