Quality Assessment of 3D Printed Surfaces Using Combined Metrics Based on Mutual Structural Similarity Approach Correlated with Subjective Aesthetic Evaluation

Abstract

Quality assessment of the 3D printed surfaces is one of the crucial issues related to fast prototyping and manufacturing of individual parts and objects using the fused deposition modeling, especially in small series production. As some corrections of minor defects may be conducted during the printing process or just after the manufacturing, an automatic quality assessment of object’s surfaces is highly demanded, preferably well correlated with subjective quality perception, considering aesthetic aspects. On the other hand, the presence of some greater and more dense distortions may indicate a reduced mechanical strength. In such cases, the manufacturing process should be interrupted to save time, energy, and the filament. This paper focuses on the possibility of using some general-purpose full-reference image quality assessment methods for the quality assessment of the 3D printed surfaces. As the direct application of an individual (elementary) metric does not provide high correlation with the subjective perception of surface quality, some modifications of similarity-based methods have been proposed utilizing the calculation of the average mutual similarity, making it possible to use full-reference metrics without the perfect quality reference images, as well as the combination of individual metrics, leading to a significant increase of correlation with subjective scores calculated for a specially prepared dataset.

Extended Damage Detection and Identification in Aircraft Structure Based on Multifrequency Eddy Current Method and Mutual Image Similarity Assessment

“…Nevertheless, although the performance of most of these improved metrics is significantly better in comparison to the “original” SSIM for many IQA benchmarking datasets, their application for some other types of images, e.g., containing multiple distortions [ 36 ] or for the surface quality assessment of 3D prints [ 37 ] not always leads to satisfactory results. Some other examples might be the development of quality metrics for audio-visual signals, stitched images [ 38 ], or light field reconstruction, compression, and display [ 39 ], where some specific types of distortions may take place.…”

“…A similar approach to mutual similarity calculation was also used for the aesthetic quality evaluation of the 3D printed surfaces [ 37 ]. However, in this case, a precise alignment has not been necessary, and no symmetry assumptions were utilized.…”

Combined Full-Reference Image Quality Metrics for Objective Assessment of Multiply Distorted Images

“…One way to do this is to design so-called combined metrics [5][6][7][8] that jointly employ several metrics (that we call elementary) in one or another way. In practice, one needs easily computable metrics and a simple way of combining them, similarly as for the 3D printed surfaces [9] or remote sensing images [10]. Because of this, the goal of this paper is to put forward a family of combined metrics that can be optimized with application to assessing the quality of images with multiple distortions.…”

No-Reference Image Quality Assessment with Global Statistical Features

“…Another line of NR-IQA algorithms focuses on combining the results of existing methods to improve prediction performance [ 32 , 33 ]. For instance, Ieremeiev et al [ 34 ] trained a neural network on the results of eleven different NR-IQA algorithms to boost performance.…”