Font Size: a A A

3D Model Retrieval Method Based On Hand-drawn Sketches

Posted on:2022-09-20Degree:MasterType:Thesis
Country:ChinaCandidate:H N ChenFull Text:PDF
GTID:2518306326983389Subject:Master of Engineering
Abstract/Summary:PDF Full Text Request
Mankind has entered the era of digital media in the 21 st century.The technologies of acquisition,storage and transmission of massive information have been developed rapidly.Although there are digital media forms such as text,image,sound and video,three-dimensional model,as a new form of digital media,has been widely used in computer-aided design,physical simulation,medical diagnosis,machinery,architectural design,education and other fields.In recent years,with the development of 3D model modeling,digitization and visualization technology,with the surge in the number of models,3D model retrieval technology has become a hot issue to be solved in various industries.Today,researchers have introduced a large number of model descriptors to describe the various features of 3D models from a visual and geometric point of view.Limited by the semantic limitations of 3D model search engine,and narrow the gap between low-level model features and high-level semantic concepts,relevance feedback strategy has become a new research field of 3D model retrieval.The development of 3D model data acquisition technology,the upgrade of graphics hardware,and the development of 3D data modeling and visualization technology have greatly promoted the development of 3D model retrieval technology.Therefore,the 3D model retrieval method based on hand-drawn sketch has become a hot research topic.In order to span the essential difference from 2D to 3D and improve the efficiency of 3D model retrieval,this paper deeply studies the difficult problem of 3D model retrieval by handdrawn sketch.In order to improve the retrieval efficiency,a recurrent neural network is proposed to dynamically recognize sketched strokes.Secondly,an algorithm for selecting the best projection view of 3D model is improved,which is used to find the most reasonable projection view of 3D model,which provides convenience for subsequent matching and retrieval work.Finally,the 3D model retrieval method based on deep learning and manifold sorting is improved.The relationship between hand-drawn sketch and 3D model projection is described with multiple constraints,and the retrieval results are obtained based on manifold sorting algorithm.Experimental results on SHREC2013 and SHREC2014 benchmark data sets show that the proposed method has better performance than the existing methods.
Keywords/Search Tags:Dynamic recognition of hand-drawn sketches, 3D model retrieval, best view selection
PDF Full Text Request
Related items