Image Similarity Assessment for Product Quality Assurance

Abstract

Image similarity is a Computer Vision technique that assesses the contextual similarity between two images. Therefore, this technique could prove useful in manufacturing quality control, where a product's image can be compared with one or a series of model and functional products' images. This research introduces and benchmarks two novel approaches utilizing the image similarity assessment process as a new approach to product quality control. The first method utilizes a twin neural network structure known as the Siamese Network useful to evaluate image similarity through their feature vector embeddings. On the other hand, the second approach takes a more traditional Machine Learning route by constructing a Table of Scores (ToS) from several image similarity metrics.