A Hierarchical Theme Recognition Model for Sandplay Therapy

Abstract

Sandplay therapy functions as a pivotal tool for psychological projection, where testers construct a scene to mirror their inner world while psychoanalysts scrutinize the testers’ psychological state. In this process, recognizing the theme (i.e., identifying the content and emotional tone) of a sandplay image is a vital step in facilitating higher-level analysis. Unlike traditional visual recognition that focuses solely on the basic information (e.g., category, location, shape, etc.), sandplay theme recognition needs to consider the overall content of the image, then relies on a hierarchical knowledge structure to complete the reasoning process. Nevertheless, the research of sandplay theme recognition is hindered by following challenges: (1) Gathering high-quality and enough sandplay images paired with expert analyses to form a scientific dataset is challenging, due to this task relies on a specialized sandplay environment. (2) Theme is a comprehensive and high-level information, making it difficult to adopt existing works directly in this task. In summary, we have tackled the above challenges from the following aspects: (1) Based on carefully analysis of the challenges (e.g., small-scale dataset and complex information), we present the HIST (HIerarchical Sandplay Theme recognition) model that incorporates external knowledge to emulate the psychoanalysts’ reasoning process. (2) Taking the split theme (a representative and evenly distributed theme) as an example, we proposed a high-quality dataset called ${\textbf {SP}}^2$ (SandPlay SPlit) to evaluate our proposed method. Experimental results demonstrate the superior performance of our algorithm compared to other baselines, and ablation experiments confirm the importance of incorporating external knowledge. We anticipate this work will contribute to the research in sandplay theme recognition. The relevant datasets and codes will be released continuously.