Incremental Image Decolorization with Randomizing Factors
Andrzej Stefan Sluzek · 2024
This paper proposes a novel decolorization algorithm that achieves high-quality grayscale outputs with exceptionally low computational cost. The method assumes, but does not apply, standard rgb-to-gray mappings. Instead, it leverages a modified floodfill algorithm with randomization. Here, the grayscale values of pixels are incrementally determined by the colors and intensities of their already processed neighbors. The randomization has a negligible impact on the image details and can help to reduce spurious artifacts commonly seen in local methods. Additionally, our approach allows seamless integration of new image fragments during or after decolorization without requiring algorithm adjustments. We evaluate our method through subjective assessments and objective performance metrics on standard benchmarks (COLOR250 and Cadik's datasets). The results demonstrate the algorithm's superiority over existing solutions in many aspects, while acknowledging identified limitations.