Automatic label placement is a crucial aspect of data visualization, essential for enhancing the clarity and readability of visual representations across various domains such as cartography, medical imaging, and emergency response management. Label, in this context, refers to textual or symbolic annotation that identifies or explains specific point feature within a visualization, such as the name of the city on a map, measurement on a medical scan, or position of AED in a schematic map for emergency response dispatchers. The work covered in this dissertation aims to advance the field by developing novel techniques addressing the inherent challenges associated with internal and external label placement in complex visualizations.   Internal label placement refers to placing labels close to the point features they describe within the boundaries of the visualization. External label placement, on the other hand, involves placing labels outside the main visualization area, connected to the relevant features by lines. 
Our research focuses on three key areas: achieving temporally stable and visually coherent boundary label placements, leveraging machine learning to improve the completeness of internal label placements, and optimizing label positioning by integrating perceptual insights. The dissertation begins with a comprehensive review of existing techniques, identifying significant gaps in handling dynamic environments and maintaining visual coherence. The literature review also highlights that the label placement quality is not entirely and precisely defined, as many cartographic guidelines rely on best practices rather than empirical studies.  Building on these insights, we introduce novel optimization methods for boundary label placement in dynamic panoramic visualizations, minimizing label movement and reducing user cognitive load. Experimental results demonstrate the effectiveness of these approaches in maintaining label stability without compromising readability or clarity.  In the context of internal label placement, we explore the relevance of deep reinforcement learning and propose a novel method that significantly improves label completeness, particularly in dense and complex scenarios.  Furthermore, we introduce a perceptual study that determines user-preferred label positions, challenges conventional placement strategies, and demonstrates the importance of considering user preferences in label placement design. Our supplementary study on users' preferred label density, a topic scarcely explored in existing literature, further confirms that integrating perceptual insights into the label placement process significantly enhances the overall user experience, leading to more intuitive and compelling visualizations. 
While the proposed methods offer substantial improvements over existing techniques, we acknowledge several limitations, including the complexity of implementing the boundary label optimization in real-time scenarios and the computational demands of the reinforcement learning approach. Future research directions include the development of mixed label placement models for 3D visualizations, optimization of computational efficiency, and further exploration of user perception to refine label placement techniques.