Silukman Image Vectorizer: A Reproducible Desktop Workflow for Configurable Raster-to-SVG Conversion

Raster-to-SVG conversion is governed by competing objectives: visual fidelity, structural complexity, output size, runtime, and operational reproducibility. Silukman Image Vectorizer is an open-source Python/PySide6 desktop and command-line workflow that integrates typed configuration, JSON-backed presets, optional preprocessing, VTracer and OpenCV-based tracing, SVG postprocessing, batch execution, and benchmark provenance. This paper describes the implemented architecture and reports a controlled benchmark over 61 provenance-recorded images from five categories. The experiment crossed two evaluated backends, three presets, and three measured repetitions, producing 1,098 planned runs after warm-up. A 4-megapixel reject policy excluded two oversized photographs across all conditions, leaving 1,062 successful evaluations and an overall execution success rate of 96.7%. Direct VTracer achieved a higher aggregate mean structural similarity index (SSIM) than the full Silukman workflow, with scores of 0.8550 and 0.8307, respectively. Within Silukman, the high-fidelity preset achieved the highest mean SSIM of 0.8990, while the low-complexity preset was fastest on average at 0.0712 seconds. Category-level mean SSIM ranged from 0.7225 for photographs to 0.9035 for logos. These observations support Silukman as a reproducible integration and experimentation workflow, not as a new tracing algorithm or a demonstrated state-of-the-art vectorizer. Generalization is limited by a small, source-concentrated dataset and a single Apple-arm64 execution environment.