| Summary: | 碩士 === 銘傳大學 === 資訊管理研究所 === 89 === As the use of the Internet and multimedia continue to grow among users, researchers need to develop a good solution for accessing multimedia data. Consequently, content-based image retrieval has become a major research with increasing important. In this study, two novel methods for content-based image retrieval in digital images were proposed. The color moment preserving technique is used in the first approach. An input image will be quantized a few colors by color moment preserving and merge similar colors to produce a few dominant colors. Then the tristimulus values and their color histograms are obtained as the color feature for image retrieval. The second approach is using color bit decomposition technique. The tristimulus values of each pixel in an input image will be decompounded into 8 bit-planes and then the sum of 1’s in each bit-planes is computed as the feature for image matching. At the same time, an additional approach tries to hybridize the both proposed methods in order to get better retrieval results. First, using color moment preserving to quantize image into a few colors and merge similar colors to get dominant colors and their frequencies as the color features, then using color bit decomposition method to extract image features from the quantized image for similarity comparison. Experimental results are given to show the feasibility and effectiveness of all the proposed methods.
|
|---|
