A novel stereo-model retrieval method in manufacturing

Abstract

With the development of computer vision technologies, manufacturing is now using stereo models for production, and the amount of stereo models generated recently has increased significantly. Thus, a fast and reliable model retrieval algorithm has become important. In this article, we propose a novel view-based stereo-model retrieval algorithm using a modified principal component analysis algorithm. Unlike a traditional principal component analysis that uses the origin of a two-dimensional image, we apply principal component analysis to extract eigenvectors for each stereo model. First, we extract a set of two-dimensional images from different directions of the stereo object. Because each two-dimensional image can be seen as one sample of the stereo object, we utilize principal component analysis to extract the eigenvectors as a dictionary for each stereo object. Then, those eigenvectors are used to rebuild the query. Finally, the reconstruction residual is applied to represent the similarity between the query and the candidate stereo object. Experimentally, the proposed retrieval algorithm has been evaluated using the ETH-80 and ALOI datasets. Experimental results and comparisons with other methods show the effectiveness of the proposed approach that can be used in engineering manufacturing and computer-aided design applications.