Consistent Image Segmentation Applied to Visual Data Mining

摘要

A novel approach for image searches and queries as applied in visual data mining applications is proposed. This new method segments an example image into real-world objects based on the Gabor Wavelet Transform. The Wavelet Transform is plotted a two-dimensional image resulting in smoothed contours for each image. The contours are processed into well-defined edges resulting in a complete segmentation of the image. The dominant colors and Gabor Wavelet coefficients are then extracted from each object and then compared with source images, which have undergone the same processing. This algorithm results in more consistent image segmentation between the example/source images thus resulting in very accurate results when comparing objects between them.