A common problem in auroral research is to find particular auroral shapes among thousands or millions of other images. Manual searching becomes impractical very quickly, which is why we are developing tools that use machine vision in analysing individual images. For each detected auroral shape in an image, a numeric representation is formed, and the comparison of shapes is performed mathematically.
This demonstration uses approximately 20,000 auroral images captured by an all-sky imager operated in Gillam (Manitoba, Canada). The images cover most of the time period between 1996 and 1998. The image analysis has been performed "off-line" ie. the objects within an image have been detected and their shapes and locations were stored in a PostgreSQL-database.
Each query, or search for similar shapes, compares the target object to all objects in the database, and returns the most similar objects. You can use any of the returned shapes to re-start another query using that shape. Naturally, if the number of objects to be compared can be limited by, for example, specifying a time interval of interest, the search will be much quicker.