Phenotype image searches are typically performed by matching text from image captions or metadata. However, the accuracy of such systems is limited because describing phenotypes with words is challenging insomuch as perception and vocabulary vary from person to person and domain expert to domain expert. How much simpler would it be if a system existed that allowed users to submit as the query an image of the phenotype they are interested in (and the system would return the most visually similar images from the database)? That is precisely the system that we been developed.
This Query by Image Example search mechanism uses a dataset consisting of leaf images from maize lesion mimic (les) mutants. To perform a search, first select an image from the box by clicking on it. A larger version of the image will be displayed. Then, hit the 'Search' button. The system will find those phenotype images whose visual appearances best match the query image.
You will also notice a set of slider bars on the page as well. These bars control the amount of emphasis placed on each class of characteristics measured from the phenotype images. For more specialized searches, adjust the sliders to weight the characteristics as desired before hitting the 'Search' button. By default, each class of characteristics is weighted evenly.
|