Metallic Dataset Creation based on FR-IQA Model for Industrial Application

As our initial investigation in Image Quality Assessment (IQA) research, the repository of image datasets for industrial applications is less than expected. There are only two primary industrial image datasets: NEU-Dataset and GC-10 DET Metallic Dataset. Both of the datasets specifically work on defect detection and image classification problem. To be precise, no image distortion was provided on the mentioned dataset. As a result, this paper aims to provide an IQA dataset image for industrial applications, especially metallic surfaces. We designed an experiment to build an industrial IQA dataset containing the real-world case of the data acquisition distortion problem, i.e., camera distortion and pre-processing image application. We made our experiment scenario based on our research assumption about the optimum distance of the data acquisition process. Thus, there are ten distortion types, and 2016 image distortions were derived from 144 reference images. To evaluate our distortion creation, we implement two FR-IQA models, Peak Signal-to-Noise Ratio (PNSR) and Structural Similarity Index Measure (SSIM). In addition, to correlate both FR-IQA models, we used Spearman Rank-Order Correlation Coefficient (SRCC) and Pearson Linear Correlation Coefficient (PLCC).