Many image classification problems can fruitfully be thought of as image retrieval in a "high similarity image database" (HSID) characterized by being tuned towards a specific application and having a high degree of visual similarity between entries that should be distinguished. We introduce a method for HSID retrieval using a similarity measure based on a linear combination of Jeffreys-Matusita (JM) distances between distributions of color (and color derivatives) estimated from a set of automatically extracted image regions. The weight coefficients are estimated based on optimal retrieval performance. Experimental results on the difficult task of visually identifying clones of fungal colonies grown in a petri dish and categorization of pelts show a high retrieval accuracy of the method when combined with standardized sample preparation and image acquisition.
|Publication status||Published - 2003|
|Event||13th Scandinavian Conference in Image Analysis - Gothenburg, Sweden|
Duration: 29 Jun 2003 → 2 Jul 2003
Conference number: 13
|Conference||13th Scandinavian Conference in Image Analysis|
|Period||29/06/2003 → 02/07/2003|