نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
Document images segmentation is one of the recent activities that have attracted researchers' attention. Unfortunately, there is no report on a benchmark dataset for Farsi document images understanding and segmentations applications that be available in the web. In the current article, a benchmark image dataset for the sake of the Farsi document images segmentation is presented, which includes 5598 images. The provided images are taken from the newspapers, textbooks and academic articles. Objects in the images are categorized and labeled into six different groups to be used easily in the subsequent applications. The object groups used in the dataset are paragraph(text), figure, table, logo, mathematical equation and header. To asset the effectiveness of the proposed document image dataset, three existing well-known methods based on deep learning are implemented on it and the results are presented.
کلیدواژهها English