In this study, we introduce a novel approach to writer retrieval systems, fundamental in handwritten document analysis, comprising feature generation and dissimilarity computation steps. Our innovation lies in the introduction of a new shape descriptor based on the Multi-Oriented calculation of Histogram Of Templates (HOT), designed to enhance the characterization of handwritten documents by incorporating orientation concepts without relying on a central pixel. Additionally, in the retrieval phase, we employ an SVM classifier based on dissimilarity learning, combining MO-HOT features with deep features. We conduct performance assessments on the CVL dataset, containing handwritten documents from 300 individuals, where our combined system outperforms several state-of-the-art systems, achieving a Mean Average Precision (MAP) of 92.89%.