Abstract
▼

As the use of digital image scanners, digital cameras and other digital devices has increased in these years, digital images in computers also have increased rapidly. In order to utilize effectively these multimedia data which are increased enormously, effective classification and retrieval techniqu...

As the use of digital image scanners, digital cameras and other digital devices has increased in these years, digital images in computers also have increased rapidly. In order to utilize effectively these multimedia data which are increased enormously, effective classification and retrieval techniques are required. In this thesis, we proposed a relevance feedback-based image retrieval method using the fusion of features. Among many features of an image, texture feature and color feature contain the basic characteristics of an image. We fused the two features into a feature and used it as the retrieval feature. First of all, to get texture feature we chose the statistic extracting method, GLCM(Gray Level Co-occurrence Matrix). We divided an image into N × M sized blocks and generated GLCM for each block. And we made the statistic contrast feature and the energy feature from GLCM of each block. To get color feature, we transformed an image into one of wavelet domain which expresses both the characteristics of space and frequency. Color feature was generated from the wavelet domain through CBA(Color Block Area) method. After generation of the color feature and the texture feature, these features were fused into a unified feature and the fused feature was applied to the image retrieval system. At last, for the better performance of the image retrieval system, we applied to the system a relevance feedback method which was based the opinions of users. In order to evaluate the precision of the proposed system, we compared the proposed method with the retrieval methods using each feature, color and texture feature. And we also evaluated the performance of the system according to the times of feedback. At the evaluation of its performance without feedback, the average precision of the methods using the fused features, the contrast of GLCM, the energy of GLCM and the color feature of CBA, are 58.85, 52.95, 56.95 and 54.47, respectively. And the method using the fused features with the 3rd-feedback gets the better performance of 3.1% than that of the method using the fused features without feedback. In the further study, we want to include other feature like shape among the image features, and expand our image retrieval system by the web-based interface

주제어

#특징 융합 귀환영상 영상검색 GLCM CBA 디지털영상;

참고문헌 (0)

1. 이 논문의 참고문헌 없음

이 논문을 인용한 문헌 (0)

1. 이 논문을 인용한 문헌 없음