Dissertation > Excellent graduate degree dissertation topics show
The Election, Alignment and Fusion Based on Static-Dynamic Multisource Features in Lipreading
Author: YangFan
Tutor: YaoHongXun
School: Harbin Institute of Technology
Course: Computer Science and Technology
Keywords: Lip-reading Characteristics of multi-source Dynamic characteristics Feature alignment Feature Fusion
CLC: TP391.41
Type: Master's thesis
Year: 2010
Downloads: 27
Quote: 0
Read: Download Dissertation
Abstract
|
Lip read lip technology as a new kind of intelligent human-computer interaction techniques in recent years is gradually moving towards practical. Lip technology, mainly in Lip identification and speaker content identification (ie, lip-reading) the two applications. This article is mainly committed to improving the performance of the latter's identification. Simple video sequences of lip region reflects the complex process of pronunciation, language processes inevitably require researchers to propose effective feature extraction method fully reflect the lip area of ??the video sequence information. However, many identity-related information contained in the video sequence. Expression of this information not only did not improve lip reading, and can easily affect the robustness and accuracy of the lip-reading problem-solving, and even lip-read information contained in the video sequence easily confused, not unified, throughout the problems in the various levels of the image structure. How the lip area video sequence contains language information fully and accurately extracted, is the starting point of this research. Lip-read information about the complexity of using different types of multi-source characteristics of the different nature to solve the problem of lip-read information described. This paper introduces the static characteristics of LBP, HOG, Gabor to describe static information on the different levels of the image. Compared to other pattern recognition problems, lip reading contains rich dynamic information. This paper presents the the rich information frame concept through the rich information frame to measure the dynamic information of the video sequence fragment accumulation, and on this basis to introduce the optical flow method of extracting the dynamic information in the video. However, the structure of the multi-source characteristic is complex, different dimensions, and not identical to the way information with structure each. Can work together to make the multi-source characteristics, it is necessary alignment. This paper presents a multi-source characteristics of the alignment of the two criteria, and two source characteristics, for example based on the characteristics of multi-source aligned, then put forward a multi-source feature alignment and integration framework, as well as two LBP, HOG, Gabor and optical flow characteristics feature fusion strategy. Finally, In this paper, the multi-source feature extraction method and the mainstream feature extraction method experimental comparison, and the experimental results are analyzed. This paper presents a multi-source alignment, fusion method is an extensible framework, no limit to the number of the characteristic features of the type and characteristics of the use of multi-source, a new idea is proposed for multi-source characteristics work together. Moreover, as long as the type of a feature and the number of features selected, it can be easily generalized to different applications in the field of pattern recognition.
|
Related Dissertations
- Feature Extraction, Selection and Combination in Lipreading,TP391.41
- Research on Methods of Fingerprint Image Segmentation,TP391.41
- Face recognition algorithm based on feature fusion research,TP391.41
- Mechanical Behavion Analysis of Free-form Surface Structure,TU399
- Research on Reentry Dynamics and Guidance of Manned Deep Space Exploration,V448.2
- Feature Extraction in Light Field Based 3D Object Retrieval,TP391.41
- The Research for Visual Surveillance System Based on Fusion of Multiple Model Data,TP391.41
- Research on the Modeling and the Grid-Connected Issue of Doubly-Fed Wind Generator,TM614
- Seismic Performance Analysis for Long-span CFST Stiff-Skeleton Arch Bridge,U442.55
- Analysis on Dynamic Characteristics and Sesimic Response of Special-Shape Concrete Filled Steel Tube Arch Bridge,U441.3
- The Construction Stages Simulation and Dynamic Characteristics Analysis of Continuous Rigid Frame Bridge,U441.3
- Dynamic Performance Analysis of CRTS Ⅲ Slab Track on Bridge,U213.244
- 1000MW Ultra Supercritical Units in the Main Steam Temperature Control System’s Characteristics and Control Strategy Research,TM621.6
- Research on Control of Supercavitating Vehicle Based on Feed Back Linearization,U661.1
- Response and Damage Analysis of Ship/offshore Platform Collision,U661.4
- Spherical fasting latticed shell static and dynamic characteristics,TU399
- Large - span continuous beam bridge construction control and optimization of closure order,U445.4
- Researching Kelly Machine & Steady with Mill in the Kelly of Manufacture,TG52
- Analysis of the Earthquake-resistant Behavior of Tall Building of Short Leg Wall with High Transfer Floor,TU352.11
- Study on Reasonable Design States and Dynamic Characteristics of Extradosed Cable-stayed Bridges,U448.27
CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile
|