Comparative Analysis of Multiple Image Vectorization Methods for SolidWorks Sketch Reconstruction and Editability in Engineering Applications

Abstract

Reverse engineering and digital manufacturing are widely applied in modern industry. The efficient conversion of two-dimensional raster images into three-dimensional CAD models is an important part of the design and production loop. However, existing image vectorization techniques generally pursue visual fidelity. When imported into SolidWorks, they often encounter defects such as loss of geometric primitives, incomplete topology, and generation of massive unordered spline curves, resulting in the complete loss of engineering editability of the reconstructed sketches. This paper proposes a quantitative evaluation system based on engineering editability. It systematically compares and analyzes the actual performance of various image vectorization methods with different underlying logics in SolidWorks sketch reconstruction. A mixed dataset containing three typical working conditions - real photos, standard engineering drawings, and hand-drawn sketches - was constructed. Four representative methods were selected for control experiments: SolidWorks Autotrace, Inkscape, Vector Magic, and an AI tool. Scores and quantitative analyses were conducted from dimensions such as the total number of entities, the extraction rate of straight lines and curve features, noise resistance, and editing stability. The results show that the native plugin can accurately extract native arcs in pure engineering drawings; the AI model has significant advantages in global noise reduction and line intention inference; the open-source algorithm can maximize the reduction of entity numbers, but gets trapped in the polygon approximation trap of converting circles into multiple lines. There is an essential conflict between the visual fidelity of the image and the engineering parametrization of CAD. There is no one-size-fits-all one-click conversion solution applicable to all working conditions. This paper specifically proposes a strategy for extracting the advantages of standard drawings and a reference method for the base image of complex images, significantly avoiding the bloated parametrization and providing an efficient and feasible technical guidance solution for engineering reverse modeling.