A comprehensive review on performance-based comparative analysis, categorization, classification and mapping of text extraction system techniques for images
Ghai, Deepika,Saxena, Sobhit
DOI: https://doi.org/10.1007/s11042-024-20257-0
IF: 2.577
2024-10-18
Multimedia Tools and Applications
Abstract:In today's quick world of images and videos, text extraction became the responsibility of machine/deep learning techniques. The various techniques utilized for text extraction reported by various authors need to be reviewed comprehensively with proper categorization and classification. The performance parameters of all available text extraction techniques need close monitoring in a variety of challenging environments to identify the best performer and also the scope for further improvement. In this work, the mapping of available techniques is accomplished with the processing steps of the text extraction system. Further, the techniques are classified on the basis of image type viz. document text, scene text, and caption text. Challenges in text extraction are segregated on the basis of various text properties available in images. For better illustration and identification of the best, detailed comparative analysis with respect to edge-based, connected component (CC)-based, texture-based, and hybrid-based techniques is presented in tabular form and bar-chart indicating performance evaluation metrics (DR%, PR%, RR%, and processing time) along with processing stage for various image types. Available datasets are summarized highlighting size, and features along with web links for providing better reach to the audience. In this work, the best-performing techniques are identified dataset-wise. The capabilities and limitations of the same techniques are discussed according to obtained parameter percentages, which provides a direction toward future work by highlighting research gaps collectively.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering