Automatic document recognition relies heavily on accurate information localization and extraction, especially in offline Arabic handwritten document processing. Many postal systems in Arabic countries, like Tunisia, still rely on manual mail sorting and processing due to the challenges of Arabic script recognition. These challenges include the inherent cursive nature of the script, context dependence of characters, segmentation difficulties, and a scarcity of deep learning-based solutions. This paper proposes a deep learning approach for segmenting handwritten Tunisian postal addresses. Our method utilizes a customized Faster Regional-Convolutional Neural Network (Faster R-CNN) architecture to achieve high detection accuracy for various postal address components.