| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 94 === Due to the progress of information technology as well as the popularity of internet and multimedia, the data amount has been increased tremendously in the last decade, where images are the most commonly used data in multimedia. In order to analyze and manage the huge image data efficiently, people need to use computer to create image files and analyze these image data. Using suitable features to describe the image data, users can browse and retrieve these image data conveniently.
The semantic message of images can assist effective retrieval of images. Therefore, the key points are how to effectively analyze the semantic message of images directly, how to encode semantic message not by retrieving image database, and how to demonstrate the meaning of images. This paper presents an approach that combines semantic-based image retrieval with content-based image retrieval in order to decrease the gap between low-level features and high-level semantic feature. Using the SOFM neural network technology, we can classify the images in database. Therefore, we can not only enhance the efficiency of image retrieval, but also can add new images into classes with only few image comparisons. The experimental results reveal that, after combining the high-level semantic feature and the low-level features of the image, the image retrieval can be improved with the classification of images.
|
|---|
