The Study of Machine Learning by using Linear Decomposition in Feature Data

• Main Authors: Yao-Chu Yang, 楊曜竹
• Other Authors: Chien-Hsiang Huang
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/65022604039242926870

碩士 === 義守大學 === 資訊工程學系 === 102 === Image content analysis techniques can effectively improve the performance of image classification, the current use of image features for image classification research, most image classification methods are based on using image comparison, comparing the extracted image feature information can effectively distinguish between similar images, but for different viewpoints or different scenes, similar content classification accuracy has to be strengthened. The purpose of the research is hoping to take advantage of machine learning methods to train, not only for image classification, but also be able to describe the image content identification dictionary. Use PCA feature reduction method to identify the composition of the main image features by using these features composition for describing the abstract meaning of image content. As per different types of image features abstract sense cluster, create a corresponding identification dictionary, the image features are able to be decomposed into linear combination of theme features, then use the coefficients of linear combination for image classification, the time required for image classification will be substantially reduced, and also be able to specify the composition of the image content. This research will use the landscape images based approach to study, use PCA to identify the composition of the image features, then use the Cluster analysis to identify theme features of each segment to create the image content identification dictionary. At last, use the identification dictionary to describe the theme features of linear combination to form coefficients of judgment as image classification.