Binarization via Local Gradient-like Differences in four Directions
Stefan Panov, Svetlana Panova · 2024
Our research focuses on a binarization algorithm tailored for standard grayscale images. An approach is proposed that computes intensity differences in four specific directions (vertical, horizontal, and two diagonal) for each pixel. These differences serve as localized gradient-like measurements, offering insights into the intensity changes around each pixel. The image is segmented into small, uniform square blocks, with each block analyzed to determine a suitable binarization threshold based on the computed differences. If a reliable threshold cannot be established, the algorithm postpones binarization for that block. In the subsequent stage, the algorithm utilizes information from adjacent blocks to finalize the binarization, ensuring consistency and smooth transitions between blocks. Additionally, the algorithm incorporates parameters that fine-tune the threshold calculation, thereby improving the binarization outcome.