Research On Algorithm Of Content-Based Image Retrieval

The traditional text-based image retrieval technology which has many shortcomings,such as heavy workload, strong artificial subjective factors, inadequate expression, the existence of language barriers in different countries, the limitations of special fields and so on. Faced with all kinds of image information which is growing at an amazing speed, the text-based image retrieval technology is far from meeting the needs of users. Content-based image retrieval based on the characteristics of the image itself, a full range of automatically or semi-automatically extracting low-level features, establishing index system,similarity matching,with the advantages of rapid, objective and efficient, which makes it quickly become a hot topic at home and abroad. This thesis is mainly about CBIR retrieval algorithm, the following is the detail work:(1) Among the global feature extraction algorithm,this thesis focuses on the research of LDP algorithm.The LDP algorithm is improved by LBP algorithm for the LBP algorithm is sensitive to the random noise and non consistency of illumination changes.In this thesis, different experiments are designed in order to verify the strong anti-noise ability of LDP algorithm,and also accomplishes the image retrieval system based on it. In order to highlight the retrieval performance, this thesis compares the retrieval results of LDP algorithm with the results of color histogram and HOG algorithm, the comparison results clearly show the superiority of the LDP algorithm.(2) Among the local feature extraction algorithm,this thesis focusing on the research of SURF algorithm. The SURF algorithm is improved by SIFT algorithm for the SIFT algorithm’s complexity of computation, easy mismatch and low real-time.Through the different conditions experiments which adds to the noise,fuzzy processing and different intensity illumination, the advantages of the SURF algorithm are proved.This thesis also accomplishes the image retrieval system based on the SURF algorithm, which get better retrieval results than SIFT algorithm.(3) For semantic gap,this thesis comes up with the SVM-based feedback retrieval,on the basic of the LDP algorithm,by learning the SVM theory and technology and combining with the requirements of this thesis,a SVM-based image retrieval system is designed.By different experiments with different parameters, we can verify that by adding the relevance feedback can really greatly enhance the overall performance of the system and can better meet the needs of uers.