Object Recognition in Herbarium Specimens

TitleObject Recognition in Herbarium Specimens
Publication TypeConference Paper
Year of Publication2011
AuthorsSteinke, KH, Gehrke M., and Dzido R.
Conference NameICDM 2011 IEEE International Conference on Data Mining
Date Published09/2011
Conference LocationVancouver, Canada
Other NumbersISSN 1864-9734
Keywordsbest template, connected component analysis, hue space, object recognition, SURF, template matching
AbstractAbout 3.5 million dried plants on paper sheets are deposited in the Botanical Museum Berlin in Germany. The aim of our project is the image analysis of scanned high resolution images. There are many different objects contained in a herbarium sheet like ruler, barcode, stamps, color-chart, labels with printed title, red type designator, envelopes, hand-written and printed annotations and the plant. Some objects are very stable like the ruler, others are extremely variable like handwriting or plant. So each class of objects requires a special detection method. Algorithms like template matching or SURF can be applied to some of the stable objects. For variable objects like handwriting detection methods are already developed by the authors and are described in other papers.
Refereed DesignationRefereed


Submitted by dpaul on

Grouping Like Objects Automatically!

In this paper the authors write, "The main aspect of this paper is the comparison of methods for localizing stable objects." Imagine, after specimen imaging, being able to use an algorithm to group all the images with a red label indicating the specimen is a type. Or, what if it was possible to group all the images in a collection that have been stamped "Imaged 2002" (or not)? How about using an algorithm to "remove the plant" from the image so that OCR sees only the text on the sheet? Check out the progress made by Steinke et al in the paper to learn more about what it is possible to do using algorithms on OCR output - to find particular objects on a herbarium sheet.