A Comprehensive Study of Intelligent Document Image Layout Analysis Using Traditional Image Processing and Deep Learning Techniques
Authors-P.Shilpa, Police Patel Radha
Keyword-Document Layout Analysis, OCR, Deep Learning, Image Segmentation, Connected Component Analysis, CNN, Graph Neural Networks, Text Detection, Skew Correction, Document Pro-cessing.
Document image layout analysis has become an essential preprocessing stage for modern Opti-cal Character Recognition (OCR) systems because the accuracy of text extraction depends heavi-ly on preserving the structural organization of document pages. Documents such as newspapers, books, magazines, invoices, historical manuscripts, and research articles usually contain complex layouts consisting of paragraphs, images, tables, figures, mathematical expressions, and multiple font styles. Conventional OCR systems often fail to maintain these structures, leading to incor-rect reading sequences and reduced recognition performance. This paper presents a detailed survey of existing document layout analysis approaches developed using both traditional image processing techniques and modern deep learning methods. Various segmentation, skew detec-tion, text and non-text separation, connected component analysis, projection profile methods, graph neural networks, convolutional neural networks, and hybrid learning models are reviewed and compared based on datasets and evaluation metrics reported in the literature. The study also examines preprocessing techniques that improve document quality before layout segmentation. A comparative analysis highlights the advantages and limitations of existing approaches while identifying research gaps for multilingual and highly complex document layouts. The survey concludes that integrating classical image processing with deep learning provides superior per-formance compared to standalone methods and offers a promising direction for future document understanding systems.
Doi-[https://doi.org/10.5281/zenodo.21735696]