Partial matching of real textured 3D objects using color cubic higher-order local auto-correlation features

Abstract

In recent years, the need for retrieving real 3D objects has grown significantly. However, various important considerations must be taken into account to solve the real 3D object retrieval problem. Three-dimensional models obtained without the use of special equipment such as engineered environments or multi-camera systems are often incomplete. Therefore, the ability to perform partial matching is essential. Moreover, the time required for the matching process must be relatively short, since the operation will need to be performed repeatedly to deal with the dynamic nature of day-to-day human environments. Furthermore, real models often include rich texture information, which can compensate for the limited shape information. Thus, the descriptors of the 3D models have to consider both shape and texture patterns. In this paper, we present new 3D shape features which take into account the object’s texture. The additive property of these features enables efficient partial matching between query data and 3D models in a database. In the experiments, we compare these features with conventional features, namely Spin-Image, Textured Spin-Image, and CHLAC features using a dataset of real textured objects. Furthermore, we demonstrate the retrieval performance of these features on a real color 3D scene.