A logo is s graphical emblem or mark used as an identification for a company and its products and services. Logos are legally protected as intellectual properties (IPs) if registered as trademarks (TMs). Logos^TM are widely distributed online nowadays in the digital economy. Due to their wide distributions online, the constant checking of TM legal usages becomes extremely challenging in the TM registration and protection system. The fact that users can easily imitate the registered TM logo designs casts serious IP legal issue, which highlights the importance of developing an automatic logo image retrieval system. Considering the complexity of TM visual semantics, this research proposes a deep embedding learning for logo image similarity analysis using triplet-network. We propose the optimization of sampling parameters to improve the TM image retrieval performance with robust model. The research aims to reduce discrepancy between human visual interpretation. This transdisciplinary engineering research incorporates deep learning (DL) modeling and TM legal analysis for image-centric TM protection. To demonstrate the model performance, more than 10,000 images for model training and 3000 images for model testing are adopted from Logo-2K+ database. Image retrieval performance shows excellent results with recall@10 exceeding 93%.