A 3D Model Retrieval Method Based on View Approximation of Minimax Game Theory

Abstract

The invention relates to a three-dimensional model retrieval method based on minimum maximum game theory view approximation. According to the new convolutional network structure, a view generator module is embedded in an existing convolutional feature extraction network, the view generator module generates views through a two-person minimum maximum game theory with conditional probability, and thefeature difference problem between the sketches and the views is solved; Then, a feature extraction layer is constructed through a residual network, and the feature separability is improved by adopting a weighting function. According to the method, the feature difference between the sketch and the three-dimensional model view is reduced, and the sketch-based retrieval accuracy is improved.

Description

technical field [0001] The invention relates to a three-dimensional model retrieval problem in the field of computer graphics, in particular to a three-dimensional model retrieval method based on view approximation of minimax game theory. Background technique [0002] With the rapid development of touch-screen devices, handheld devices such as mobile phones and tablets are widely popular due to their good portability, but this also brings about human-computer interaction problems with related devices; on the other hand, with the advancement of virtual reality technology , the research on the retrieval task of 3D models has also attracted more and more attention in the fields of computer vision and computer graphics. Because sketches can directly express people's thinking and purpose, using sketches to retrieve similar 3D models has become an important research direction. The problem of 3D model retrieval based on sketches is to retrieve similar 3D models for each hand-drawn...

AI Technical Summary

Problems solved by technology

But the traditional method based on the three-dimensional model retrieval method of sketch is less effective, such as CN101004748A at first obtains the view database by the three-dimensional model database, and obtains the feature database by the view database; Match the features in it, calculate the similar distance between the two-dimensional sketch and the three-dimensional model, and sort the similar distance; finally return the position, index image, URL, etc. of the top three-dimensional model; the invention CN103177098A discloses a hand-drawn A three-dimensional model retrieval method for graphs, the invention first generates multi-view contour graphs, and extracts corresponding occupancy map features, distance transform features, contour signature features, Fourier descriptors, Hu moment features and Poisson features; then all multi-view Combining the contour features of the angle of view to form a new feature of the corresponding dimension, similarly, it can also form a new feature of the corresponding hand-drawn image; finally, through the k-d tree feature matching method, find out the new feature of the hand-drawn image that is most similar to the feature of the 3D model outline Features; the invention reduces the sensitivity of parameter settings, thus improving the retrieval effect

Images