3D Model Retrieval Method Based on Deep Convolutional Neural Network

Abstract

The invention discloses a three-dimensional model retrieval algorithm based on a deep convolutional neural network. This method uses a metric learning algorithm to obtain a Euclidean embedding space, embeds hand-drawn sketches and model projections in the same feature space, and the Euclidean distance in the feature embedding space It can directly represent the similarity between the sketch and the model projection, solving the cross-domain matching problem between the sketch and the model projection. At the same time, a sorting mechanism is designed so that the distance between images of the same category in the feature space is smaller than the distance between images of different categories, which can distinguish subtle differences between different categories and adapt to variants of different styles in the same category; and the present invention The convolutional neural network is used to learn an ultra-complete feature filter group to form a feature extractor to extract high-level abstract features, which effectively solves the problem of weak generalization ability of hand-designed low-level geometric feature descriptors and difficulty in extending to unknown data sets. The problem.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and relates to a three-dimensional model retrieval method based on a deep convolutional neural network. Background technique [0002] At present, 3D models have been widely used in fields such as virtual reality, industrial design, 3D games and visual design. With the development of 3D graphics modeling technology and 3D data acquisition technology, massive 3D model databases have been produced. Therefore, in order to make full use of the existing 3D models and help users obtain 3D models that meet their needs conveniently and efficiently, 3D retrieval technology has become a current hot research issue. [0003] The workflow of 3D model retrieval is to search out the relevant model collection in the model database according to the query request input by the user, and finally feed it back to the user. One class of algorithms uses existing 3D models as input to express query intentions, b...

AI Technical Summary

Problems solved by technology

And the present invention adopts the convolutional neural network to automatically learn a super-complete feature filter group to form a feature extractor, and extract high-level abstract features. In addition to being able to tolerate nonlinear deformation, this feature also has a strong generality in unknown data sets. It effectively solves the problem that the hand-designed low-level geometric feature descriptor algorithm has weak generalization ability and is difficult to extend to unknown data sets.

Images