Document text detection is a key step in document analysis and optical char-acter recognition for its importance on automatic content extraction, making it more readily accessible. Traditional methods, relying on preprocessing and feature extraction, often struggle with varied layouts and text orientations. This paper introduces an FCNN-based approach for detecting text in documents, leveraging an end-to-end framework that predicts text regions rather than quadrilateral bounding boxes. The FCNN model adeptly handles diverse text orientations, fonts, and layouts through region proposals and semantic segmentation maps. Evaluated on benchmark datasets, including synthetic and real-world document images, our method outperforms state-of-the-art techniques, demonstrating high accuracy and robustness in detecting text across various document structures. This advancement enhances document analysis and data extraction systems.