spark
Project home

Two versions of SPaRK are available. The difference is the clustering algorithm used for automatic date/time segmentation. One uses a nearest neighbor algorithm and local mean as the threshold for event detection and the other uses a sliding window algorithm with a frame size that calculates an average for the threshold.

SPaRK is a system to organize images using location, date, event annotation, and semantic noun relationships via WordNet. SPaRK uses an automated date time event clustering algorithm to segment the images into date time hierarchies; these segments are further grouped and bound by a location. An annotation can then be applied to the event timespan which introduces a semantic hierarchy used to automatically locate related events in the digital image collection.

SPaRK uses existing technology in order to redefine the methods associated with organizing digital images. Other works that use location data, date time and other attributes such as time of day to organize information exclude an important facet of cognition, the semantic meaning.

SPaRK directly addresses this requirement by incorporating a lexical memory component to help facilitate the automatic clustering of images into events. SPaRK uses this additional information to find related events that span both spatial and temporal differences. These related events can give rise to special correlations such as cause and effect given that location and date time hierarchies are established before the semantic relationships are formed.