Summary: | 碩士 === 元智大學 === 電機與資訊工程研究所 === 87 === In this thesis, a coarse-to-fine hierarchical classification approach based on the features derived from cellular color decomposition is proposed. In color quantization step, since HS structure of HSV color model is a hexagon that as the same as cellular pattern, cellular decomposition is used in the proposed method to cluster HSV color space. In classification step, five image-based features extracted directly from the quantization results are used to prune irrelevant database images in coarse stage. In fine stage, two cluster-based features are extracted from closest-cluster matching results. Obviously, these features are extracted only from the small set of plausible images such that classification process can be speeded up. According to cluster characteristics, class-based similarity measure is proposed in this study to evaluate image to class similarity. Candidate images are then ranked on the basis of the combined similarity measure derived from individual-based similarity measure, in terms of one image-based feature and two cluster-based features, as well as class-based similarity measure. Finally, K-NNR is used to classify the query image into one class according to ranking results. The proposed approach is compared with different methods in experiment. Various experimental results prove the effectiveness and practicality of the proposed method.
|