Optical Character Recognition (OCR) is a widely utilized technology that converts text from images or handwritten documents into digital formats. However, recognizing handwritten and printed text, as well as text within images, remains a complex challenge due to variations in writing styles and the intricate nature of characters. This study presents an innovative OCR method using an Encoder-Decoder LSTM framework that integrates CNNs) and RNNs. The proposed EDLSTM architecture is designed to independently extract features from raw image data and identify sequential text patterns. It incorporates seven convolutional layers and two LSTM layers, enabling effective recognition of text embedded in images with complex backgrounds and diverse fonts. The system demonstrated exceptional performance across multiple benchmark datasets, highlighting its ability to enhance the accuracy and efficiency of text recognition tasks. A comparative analysis of models revealed the accuracy levels achieved: CRNN (78%), attention-based RNN (88.5%), DenseNet201 (87%), and EfficientDet-D4 (96.5%). Remarkably, the proposed Encoder-Decoder LSTM model outperformed these with an accuracy of 98.32%, establishing it as a state-of-the-art solution in OCR. This research provides a thorough investigation into OCR methodologies using EDLSTM, emphasizing its potential for advancing text recognition technologies.