3D Model Retrieval Method Based on Optimal View and Deep Convolutional Neural Network

Abstract

The invention relates to a three-dimensional model retrieval method based on an optimal view and a deep convolutional neural network. The invention first extracts views of a three-dimensional model from multiple viewpoints, and selects an optimal view according to the order of gray entropy. Secondly, the view set is trained through a deep convolutional neural network to extract the deep features of the view and perform dimensionality reduction. At the same time, the edge contour map is extracted from the input natural image, and a set of 3D models is returned after similarity matching. Finally, the result list is fine-tuned and rearranged based on the proportion of the category of the target model in the retrieval results, and the final retrieval results are returned to realize 3D model retrieval. This method effectively selects a better view, reduces view redundancy, and uses deep features to The view is expressed at a higher level, which effectively improves the retrieval effect.

Description

technical field [0001] The invention relates to the field of computer graphics, in particular to a three-dimensional model retrieval method based on an optimal view and a deep convolutional neural network. Background technique [0002] With the increasing computer graphics processing capability and 3D modeling technology, 3D models have been widely used in many fields such as industrial design, virtual reality, and medical diagnosis, and the number of 3D models has also grown explosively. Massive data brings new opportunities and challenges to the development of 3D model retrieval technology. [0003] In the field of view-based 3D model retrieval, the common research idea is to render the 3D model into multiple 2D views, extract the artificially designed features of the views, and then match the features of the input source image with the view features to achieve similarity matching. Get the target model. However, this search method has the following two problems: [0004...

AI Technical Summary

Problems solved by technology

[0004] (1) In the existing view selection methods, researchers usually choose to surround the 3D model with a polyhedron, use the vertices of the polyhedron as the viewpoint for view selection, and extract views consistent with the number of vertices, or use grid saliency, visible Area ratio, maximum area projection method and other methods to select the optimal view, but the quality of the viewpoint obtained by the former is different, and the extraction effect of the latter for different types of models is quite different

Images