Font Size: a A A

The Technology Of Content-based Image Retrieval Research And Design

Posted on:2011-03-20Degree:MasterType:Thesis
Country:ChinaCandidate:E S YanFull Text:PDF
GTID:2178360305482280Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
The main idea of CBIR(Content-based Image Retrieval) is that, for a given image, features are extracted by some algorithms and based on the similarity measure function.Its characteristics are used to match the features of the image database. According to the the high and low matching degree, retrieved images are convenient for user to select. In this process, feature extraction and similarity measurement are the most critical.For the characterization of the contents of an image can be directly affected by feature extraction, and higher real-time the calculation is required by similarity measurement. Both of them are focused by this paper.Firstly, the image retrieval of the origin is analyzed in this paper, and then the hot spot at home and abroad is elaborated to set our research direction. Some key techniques of the image retrieval are studied in detail, such as color feature, texture feature, shape feature.The selection of similarity measurement function, different measurement methods and feature extraction methods are choosed by those different characteristics. Secondly, the retrieval system must be designed and conducted the experiments after selecting the appropriate feature extraction and measurement methods.And the experiment's result must be judged by their effects through precision rate and recall rate. The main target of this paper can be divided into three parts:the color feature and texture feature are focused in the first part.For the color feature, edge detection and color histogram are used to feature extraction.In this course,edge detection is used to determine interesting areas and then calculate its color histogram.After that,European distances is used for similarity measure. Compared to a single color feature, the experimental results show that accuracy is increased.So the purpose of experiment is acheived. The second part is for texturefeature, image is transformated into gray scale of image, and then gray co-occurrence matrix is generated. Five feature vectors including energy, entropy, inertia distance, local equilibrium, correlation are calculated through the matrix.European distance measurement is also used for similarity measurement.The image which'has obvious texture features is used to conduct the experiment. Compared with color feature, retrieval effectiveness is much better. The third part is focused on the two characteristics of the search results which are improved.As a single feature,the effect is limited.So we propose our schem:integrated more characteristics image retrieval. The texture and color characteristics are considered to improve the search results.As traditional feature extraction method is used in this paper, retrieval effectiveness is limited because of the image content is too complex and boundaries between different features are not obvious. Some new method, such as relevance feedback and semantic retrieval of relatively are proposed by researchers, owning to lower experimental level. No breakthroughs are made, so a long way must be experienced to further our research.
Keywords/Search Tags:Image retrieval, Feature extraction, Similarity measurement, Histogram, Co-occurrence matrix
PDF Full Text Request
Related items