Improving Image Tracing with Artificial Intelligence
Andreas Fischer, Manuel Amesberger · 2021
Image tracing describes the task to convert a raster image into a vector format. This paper investigates whether dimensionality reduction techniques can improve the results of image tracing by producing simpler vector graphic files while keeping image quality reasonably high. In particular, an Autoencoding Neural Network and Principal Component Analysis are investigated as pre-processing steps before performing the actual tracing. Results indicate that using an Autoencoder as pre-processing step before image tracing can reduce SVG file complexity by over 70% with acceptable impact on image quality.