Stereo vision three-dimensional human face modelling approach based on dummy image

Abstract

It uses a 3D model of reference human face (HF) to estimate the HF gesture parameters in the input stereo image (ISI) pair to generate the virtual image (VI) pair of reference HF (RHF) under the same gesture. Due to the corresponding between VIs is known, the corresponding is extended from VI to ISI via calculating the correspondence between the input HF and RHF under the same gesture. This greatly improves the matching effect of HF stereo image pair and ensures the HF point-cloud (PC) data more accurate obtained from calculating stereovision difference. This method converts 3D PD optimizing problem into a 1D curve optimizing one to greatly raise processing speed. It finally builds HF 3D surface from the optimized PC, then texture-mapping it to gain a HF 3D model (H3M) with a high true sense. This invention can meet various requirements against H3M under various practical application situations. It is practical, fast, accurate and has a extensive application foreground.

Description

Technical field: [0001] The invention belongs to the technical field of image processing and computer vision, and in particular relates to a method for establishing a three-dimensional model of a human face by applying stereo vision. Background technique: [0002] Currently, image-based 3D face modeling methods have been extensively and deeply researched. "Computer Application" magazine, 2000, Vol. 20, No. 7, pages 41-45 and 255-257, proposed a method for generating a three-dimensional model of a human face using two orthogonal face images of the front and side. The disadvantage of this method is that the input The orthogonality of the image has strict requirements. Chinese Patent Application No. 200610088857.9 proposed "Fast Face Modeling Method and System Based on Single Photo" and 200610024720.7 proposed "A Method for Realizing Fast Face Model Reconstruction from a Frontal Image" to simplify this method, only The frontal image is used for 3D face modeling, but because t...

AI Technical Summary

Problems solved by technology

Chinese Patent Application No. 200610088857.9 proposed "Fast Face Modeling Method and System Based on Single Photo" and 200610024720.7 proposed "A Method for Realizing Fast Face Model Reconstruction from a Frontal Image" to simplify this method, only Use the frontal image for 3D face modeling, but because it does not obtain the real depth information of the specific face, the generated 3D model of the specific face is not very accurate

However, there are many redundant information in the video sequence, and the computational complexity is high.

The Morphable Model proposed in "IEEE Transaction on Pattern Analysis and Machine Intelligence", Volume 25, Issue 9, Page 1063-1074 in 2003, adopts the human face three-dimensional database The samples are learned to guide 3D face modeling, but this method requires a large amount of 3D face data as the basis, which is difficult to meet in practical applications

"Image and Vision Computing" (Image and Vision Computing) 2000, Volume 18, Issue 4, Page 337-343, proposed a 3D face modeling method based on stereo vision. The large ambiguity makes the generated 3D face model very unsatisfactory and greatly deviates from the target face. Therefore, it is necessary to introduce a lot of constraints and prior conditions to optimize the model in 3D space, which increases to a large extent. the complexity of the modeling work

Images