| 摘要 |
The present invention relates to tensor-voting-based color-clustering, and more particularly, to a tensor-voting-based color-clustering system and method which involve automatically estimating the number of dominant colors in tensor voting during the performance of color-clustering of an input image, and perform clustering using the data density of a color, thereby achieving effective image segmentation. To achieve the above-described technical features, provided is a tensor-voting-based color-clustering system comprising: a color feature space generating unit for classifying colors in an input image to generate color clusters for each color; a tensor voting unit for casting a tensor vote for each color cluster generated by the color feature space generating unit; and a color space analyzing and segmenting unit for analyzing and segmenting each color cluster for which a tensor vote is cast by the tensor voting unit. In addition, to achieve said technical features, provided is a tensor-voting-based color-clustering method using the tensor-voting-based color-clustering system comprising the color feature space generating unit, the tensor voting unit and the color space analyzing and segmenting unit, wherein the method comprises: a color feature space generating step in which the color feature space generating unit classifies colors in an input image to generate color clusters for each color; a tensor voting step in which the tensor voting unit casts a tensor vote for each color cluster generated through the color feature space generating step; and a color space analyzing and segmenting step in which the color space analyzing and segmenting unit analyzes and segments each of the color clusters for which a tensor vote was cast. The system and method of the present invention having the above-described objectives and functions may improve the image segmentation of an object as compared to conventional techniques by performing color-clustering. Moreover, the system and method of the present invention enable the number of primary colors to be automatically estimated unlike in conventional techniques, thus rendering the result of each color-clustering uniform and stable. |
