An Archdaily Lexicon is a search engine of architectural images. It challenges the representational paradigm suggested by the Building Information Modeling where everything has to be represented using the same pre-defined ontology.
An Archdaily Lexicon proposes to encode objects by their relations to the plenty - a finite symbolic set from which objects are 'spelled' as of spelling texts with an alphabet. The symbolic set is derived from the plenty and carries no semantic meaning.
An Archdaily Lexicon demonstrates this concept by clustering image patches of all ArchDaily images. The obtained clusters encode a given image by their Euclidean distances to the patches of the image. The query and comparision of the images are therefore based on the similarity of the encodings rather than the pixel colors.
Symbolic set as the visual elements
Image encoding using the symbolic set
case 1. "text" to "text": search for similar images
Similar image search by the input source (top left)
Similar image search by the input source (top left)
Similar image search by the input source (top left)
case 2. "words" to "text": search for images that contain input objects
Patch-to-image search by the input patches (top left)
Patch-to-image search by the input patches (top left)
Patch-to-image search by the input patches (top left)