Line Segmentation in Persian Texts in Double Columns Using Hierarchical Clustering Algorithms

This study addresses the line detection problem in digital images containing Persian text, which presents unique challenges due to the language's specific structural characteristics. The challenges include line deviations, overlaps, varying fonts, and the presence of unique symbols in Persian scripts. The proposed method utilizes a combination of preprocessing techniques, connected components analysis, and the DBSCAN clustering algorithm to effectively separate text lines from two-column images. By applying Gaussian filtering, morphological operations, and density-based clustering, the method successfully detects and isolates individual lines of text. The results demonstrate high accuracy, achieving 98.31 % in line detection across a dataset of 100 images. The proposed approach illustrates significant potential for improving document management and text recognition within Persian language processing.