Binarization of grayscale images using the local threshold filtering method

The article considers the urgent problem of halftone image binarization, which plays a key role in computer vision and pattern recognition systems. Traditional binarization methods based on global threshold values demonstrate significant limitations when processing images with uneven illumination, noise and complex texture structure. The purpose of the study is to develop and theoretically substantiate a local threshold filtering method that takes into account the spatial characteristics of the processed image areas. The proposed approach is based on the adaptive selection of the threshold value for each pixel depending on the statistical parameters of its neighborhood, which allows dynamically adjusting the binarization criteria. A distinctive feature of the developed method is the use of multi-scale analysis of structural elements of the image with subsequent optimization of local thresholds based on minimizing the error functional. The experimental verification of the algorithm on a set of test images of varying complexity demonstrated an increase in binarization accuracy by 12…18% compared to known methods, which confirms the promise of its application in medical image segmentation, document analysis and industrial quality control systems.