Research On Facial Feature Extraction And Matching Algorithms For Image Retrieval

With rapid development of information technology, people require more andmore for information acquisition. Information retrieval has already beyond the tradi-tional text media, and recently the content-based multimedia retrieval which is rep-resented by image retrieval becomes the research hotpot. The content-based imageretrieval technology possesses obvious advantages in description and automatical in-dexing capability, while there are still many problems to be solved corresponding tothe text-based retrieval technology. Using the facial image which is a common subsetof image as a breakthrough point, this thesis mainly concentrates on the content-basedfacial image retrieval technology, which aims to propose suitable facial feature extrac-tion and matching algorithms for retrieval background, moreover to narrow the gapbetween theory and practice so as to promote the development of the image retrievaltechnology.Base on previous target, this thesis makes emphasis on facial features extractionand matching technology which is applied in retrieval background. This work con-tains statistical features research and geometrical features research. More specifically,introduced the facial image retrieval algorithm based on statistical and geometricalfeatures under the single query sample condition, and verifies the system frameworkof content-based facial image retrieval. The main research work of this thesis is asfollows:Give analysis and comparison between the traditional face recognition sys-tem model and the facial image retrieval system model. Put forward a novel facialimage retrieval system model which fits the requirement of the facial image retrievaltechnology well.Promote four single sample enlargement methods, which considering the tech-nical requirement of the facial image retrieval system model to the statistical featureextraction and matching algorithms, so that generate the enlarged sample set of thequery image, and also compare and conclude four single sample enlargement meth-ods. recognition field, to extract and match statistical features in order to complete theprocess of facial image retrieval. Additionally, sufficient system retrieval results fromdifferent combination among single sample enlargement methods and statistical fea-ture extraction and matching algorithms are concluded.Induct geometrical features in facial image, concise suitable geometrical fea-tures and propose the facial image retrieval scheme based on geometrical features.Realize face detection and facial feature location by using Adaboost algorithm andAAM algorithm respectively. Furthermore predict the possible usage of geometricalfeatures in the facial image retrieval system.Follow the software engineering theory, analysis, design and implementationare used to complete the demonstration system for the facial image retrieval. Eachsoftware model is developed according to reusable thought and technology. Interfacesfor enlargement methods, training algorithms and retrieving image data-set have beenreserved, so system integration and extension become feasible.