Research on content-based retrieval in image data bases has focussed on searching image properties such as color, texture, histogram, and shape. We have made considerable progress in developing algorithms for texture-based search [10]. We are currently investigating augmenting the ADL catalog with indices based on texture features.
The basic idea is to
extract texture information from the images as they are ingested.
This is done using Gabor filters,
which are modulated Gaussians. Processing through a bank of these
Gabor filters is approximately equivalent to extracting line edges
and bars in the images, at different scales and orientations. Simple
statistical moments, such as the mean and standard deviation of the
filtered outputs, can then be used as indices to search the database.
Fig. 3 
shows an application to browsing large air photos.
Figure 3: Browsing a large air photo using Gabor texture features.
The figure shows a downsampled (reduced-resolution) version of the image on the left, and the retrieval results using the query pattern in the first column. The query pattern was selected from the region containing some buildings in the left center of the air photo. The retrieved subimages, ordered according to a similarity measure (top to bottom and left to right in the last two columns), are all from the same region.
For the WP, we plan to create a database of air photos which can be searched using texture templates. Texture information will be extracted when the images are ingested. A small set of texture templates will then be created which represent the different textures that may occur in the air photos. A user initiating a search can chose an image region (by pointing the cursor) and the region's texture will be used to retrieve matching texture templates. Each of these texture templates will have pointers to the air photos where they occur.
Instead of Gabor filters, one may also use the same orthogonal wavelet transform that was used for storing the image data. However, experiments on a large set of textured images have shown that the retrieval performance of conventional orthogonal wavelets is not as good as that of Gabor filters [11]. Unfortunately, Gabor transforms are awkward for storage applications. In particular, they do not form an orthogonal basis set. Many researchers have used Gabor transforms for image compression, mainly for lossy compression. However, no efficient algorithms exist for computing the forward and inverse transformations, which is important in a digital library context. While data ingest is off-line and can be computationally intensive, data retrieval should be both fast and performed in real time using existing hardware. Orthogonal wavelets are good for such implementations, whereas non-orthogonal Gabor wavelets are good for image analysis.
