Effectiveness of Wavelet Transform, Fractals, Lacunarity and Spatial Indices in Content-Based Image Retrieval
Date of Award
Master of Arts
Dr. Charles Emerson
Dr. David Dickason
Dr. James Biles
Masters Thesis-Open Access
The thesis demonstrates the feasibility of applying wavelet transforms, fractals, and spatial indices in content-based image retrieval (CBIR) of remotely sensed imagery. It consists of three major parts: the design and development of thick-client CBIR software, population of two samples of SPOT™ and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) image databases with metadata describing their visual properties, and, exploration of the effectiveness of region quadtree query and retrieval by content using various combinations of indices. 272 SPOT™ and 507 Landsat 7 ETM+ 512 x 512 pixel panchromatic images of Atlanta, Georgia were subsetted from two scenes. Moran's I index of spatial autocorrelation, two sets of fractal dimensions (based on the triangular prism and box counting methods), lacunarity, 5-bin spectral histogram and wavelet energy signatures (12 indices at 3 levels) were computed at each step of the six level region quadtree and the results were stored in a relational database. To query the populated database, region quadtree was used to select features of interest within a randomly selected image. Associated indices of the selected quadrants of the query region were computed and compared to the metadata in the database. Individual applications of each index on the whole image was found to reveal a promising pattern but picking the right indices and combining them with equal weights seem to pose a challenge especially when smaller quads are used.
Chinniah, Sivagurunathan, "Effectiveness of Wavelet Transform, Fractals, Lacunarity and Spatial Indices in Content-Based Image Retrieval" (2005). Masters Theses. 3921.