Pattern-Recognition-Driven Autonomous Painting Generation: A Framework for Creating Images Through Learned Drawing Actions

Most contemporary image-generation systems represent image synthesis as the transformation of a latent representation, noise distribution, or learned feature representation into an image. An alternative is to formulate image generation as an autonomous painting process in which an artificial agent constructs an image through a sequence of drawing actions. Although stroke-based rendering, reinforcement-learning-based painting, and neural painting have been investigated, an important research question remains: how can pattern recognition actively guide the agent's own creative decisions throughout the entire painting process, rather than merely guide the reproduction of a predefined target image?

This paper proposes Pattern-Recognition-Driven Autonomous Painting Generation (PRAPG), a conceptual framework in which an agent begins with a blank canvas and progressively constructs an image through learned drawing actions. The agent recognizes emerging visual patterns, estimates the current state of the composition, identifies underdeveloped or inconsistent regions, generates candidate artistic intentions, selects drawing actions, and evaluates their consequences. The resulting image is therefore constructed through explicit painting operations rather than directly synthesized as a final pixel array. The proposed framework combines visual pattern recognition, hierarchical composition planning, action-based rendering, uncertainty-aware action selection, and iterative self-evaluation. A central hypothesis is that integrating recognition into the control loop may improve structural coherence, controllability, and interpretability compared with painting systems that primarily optimize stroke parameters against a fixed reference image. This paper formalizes the framework, presents a learning objective and training strategy, and proposes an experimental protocol. It does not claim empirical validation or establish that the framework is superior to existing approaches.