Cursive writing poses a challenge for OCR (optical character recognition) systems, making it difficult for them to accurately process. Traditional methods struggle with this task, so new and innovative approaches are necessary to enhance their performance. This study introduces a method by incorporating an RPN (Region Proposal Network) into OCR frameworks. The RPN, which is based on learning principles, identifies and evaluates areas of interest in an image. By integrating this network, the goal is to improve the detection and recognition of handwriting within OCR systems. Furthermore, adjustments to the loss functions have allowed the proposed technique to achieve accuracy in training sessions while reducing regression loss. Moreover, the system achieved an accuracy of 87.27% using VGG and preprocessing methods, showing improvement compared to the 71.25% accuracy of the CNN model paired with Bi-LSTM without optimization. This research highlights the effectiveness of RPN model-based strategies in interpreting handwriting within OCR frameworks. Our results indicate that with adjustments and enhancements, OCR systems can effectively tackle the complexities associated with writing styles, thereby broadening their capabilities and usability.