A Review on Text Detection and Recognition in Images and Videos Using Machine Learning and Deep Learning Techniques

Reliable text detection and recognition systems are needed in order to enable efficient indexing, searching, and semantic retrieval due to the increasing volume of digital video and image content. For extracting and recognising textual information from video frames and images, considerable research has focused on Optical Character Recognition (OCR) techniques since long. The aim of this review is to study the evolution of various methods for text detection and recognition, tracing the transition from traditional feature-driven approaches, such as the Haar Discrete Wavelet Transform (HDWT), to modern machine learning based architectures. The objective is to support ongoing research in regional language text detection and recognition. Text analysis of the Gujarati script presents unique challenges, including its multi-zonal character structure and the absence of a shirorekha (header line), which distinguish it from scripts such as Devanagari. Additional complications emerge from elaborate character shapes, conjunct forms, diacritical markings, and considerable structural variety. Early study demonstrated the efficacy of wavelet-based methods for text localization in videos and images, setting the stage for later studies. Recently, sophisticated models employing Convolutional Recurrent Neural Networks (CRNN) and Long Short-Term Memory (LSTM) networks have achieved state-of-the-art performance, demonstrating enhanced resilience to intricate backgrounds, fluctuating lighting, and the cursive features of Gujarati text in dynamic visual environments. This study aims to advance the field of regional language text detection and recognition in images and videos.