| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 103 === This paper discussed image abstraction based on bilateral filter and Gaussian mixture model (GMM), respectively, and proposed methods for optimizing the parameters and improving image quality. As regard the methods based on bilateral filter, the entropy of the image is proposed as an indicator for determining the color variance of bilateral filtering, though which better abstraction results could be obtained. In addition, the color variance of the bilateral filter could be further tuned dynamically according to the distance between each pixel and its closest edge point. This may alleviate effectively the problem of color mixing, that the colors around the boundaries of adjacent areas tend to be mixed during bilateral filtering and improve the quality of abstracted image. For the methods based on GMM clustering, the combinations of color features and spatial features are studied first. Furthermore, a method of selecting the number of clusters according to the average quantization error of color was proposed. It was shown that this method can achieve better abstraction results because the clusters may contain the major colors of the image with fewer distortions. Finally, a fuzzy clustering approach was utilized to eliminate the unnatural breaks around the boundaries of the clusters, and the results with better visual quality can be obtained.
|
|---|
