Topological and Perceptual Representations for Reference Frame Retrieval in Hand-Drawn Animation

Abstract

Hand-drawn animation is a resolutely human art form, with its unique artistic and commercial appeal emerging directly from the emotion and soul captured within each of the 100,000+ hand-crafted frames of a feature-length film. Remarkably, watching hand-drawn animation invokes the "illusion of life" within the human mind, whereby highly stylized characters formed from collections of simple 2D shapes are somehow perceived as real 3D entities, despite their penchant for physically impossible motion and bodily deformations. Although this perceptual phenomenon has not been widely empirically studied, the small yet vibrant field of hand-drawn animation technology research aims to computationally model its key elements, such that frames' material collections of 2D lines and shapes are automatically interpreted with respect to the "living" characters they depict.

Despite the extensive and complex technical challenges hand-drawn studios face during animated film production, many of the technical solutions produced by hand-drawn animation technology researchers are rarely used by high-level animation studios. This fact reveals a disconnect between technology researchers' and professional animators' respective understandings of technology's role within the hand-drawn animation process. Our research seeks to gain insights into and begin to resolve this disconnect.

This thesis begins by evaluating the present state of hand-drawn animation technology research through a novel analytical lens. We consider the field subject to its deep theoretical roots in computational topology, aided by artistic and practical insights from hand-drawn animation industry experts. Our analysis indicates that technologies' artistic advantages and limitations arise directly from their chosen topological approaches, with topological sophistication generally improving artistic functionality. Crucially, advanced topological approaches to hand-drawn animation technology appear to emulate cognitive processes likely underpinning hand-drawn animation perception.

With this in mind, our research pursues the first application of Topological Data Analysis (TDA) to hand-drawn animation technology, focusing on advanced topological frameworks' potential to model human visual perception of hand-drawn animation. Advised by hand-drawn animation industry experts, we focus on the specific problem of computationally modeling animators' Reference Frame Quality (RFQ) perception -- that is, how good one hand-drawn frame is as drawing reference for another. To this end, we conduct perceptual experiments on professional animators at the hand-drawn studio Cartoon Saloon, each exploring how animators' specialized visual perception skills practically impact different aspects of RFQ perception.

From our perceptual findings, we construct a novel general-purpose TDA framework for complex system analysis, rigorously defining its mathematical, conceptual, computational, and data analytical facets. To demonstrate our framework's efficacy, we detail its application to perceptually-based RFQ analysis, illustrating its suitability for modeling hand-drawn frame perception. In future, our framework will underpin a reference frame retrieval system for use within hand-drawn animation studios.