The technology analyses whether text is contained in images based on the graphic pattern of words and lines, said developer Eugene Smirnov.
The method is able to recognise text in almost any language and is not affected by techniques such as warping that spammers use to avoid detection, he said.
The system determines whether or not detected text is spam by comparing it to the contents of a database of spam templates.
Spam is expected to continue to be a problem in 2009, particularly with the rise in the number and popularity of websites that allow user-generated content.
Cyber criminals are expected to move to a more distributed way of controlling and hosting malcode after two main criminal spam hosting companies were shut down in 2008.