Efficient Text Extraction from Product Images Using Deep Learning and Parallel Computing

The domain of deep learning, particularly in the context of text detection and recognition, has witnessed remarkable progress over the years. Text detection and recognition entail identifying and extracting textual information from images, an essential component in various real-world applications. The ability to extract text robustly and efficiently from scenes is essential for interpreting traffic signs or content-based image retrieval. This domain has been greatly influenced by the advent of Conventional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), which have demonstrated a superior capability to handle diverse text shapes and irregularities. The utilization of these models has opened new horizons for text detection and recognition, allowing for a more flexible approach to accommodate the wide range of text forms found in the real world, such as curved or skewed text. Despite significant progress in the field, performance challenges persist, notably the time-consuming nature of text extraction from images. As data volumes grow, the need for faster extraction becomes increasingly critical. Existing methods may not fully harness the potential of parallel computing. Addressing these issues is essential for advancing text detection and recognition for practical applications, which is the focus of our research. We implemented parallel text extraction using the Optical Character Recognition (OCR) engine within Kaggle Environments, significantly improving efficiency. The parallel implementation processed text extraction 6 times faster than the sequential approach.