On Extracting Perception-Based Features for Effective Similar Shader Retreival

A similar shader retrieval searches for shaders similar to a given query shader, and significantly reduces trial-and-errors and long processing time in a shading process. However, the developing of similarity measure is quite challenging because of the two characteristics of shader: (1) Shaders have different numbers of attributes that are peculiar to each of them, (2) Since the number of attributes in a shader becomes up to hundreds, the ´dimensionality curse´ occurs. In this paper, we propose a novel method for extracting perception feature in effective similar shader retrieval. The proposed method finds low-dimensional features by analyzing hundreds attributes in a shader. The characteristics of extracted features are exactly the same with all the shaders, thereby making the similar shader retrieval much simpler. The proposed method constructs the perception features representing a shader by analyzing the meanings and relationships of all the attributes in the shader. Therefore, the method provides accurate results in the similar shader retrieval. To show the effectiveness of the proposed method, we conduct a variety of experiments.