Stockman MSU CSE1 Image Database Access Find images from personal collections Find images on the web Find images from medical cases Find images from art collections Find images from architectural cases
Stockman MSU CSE2 General methods of query Use prestored symbolic keys – standard Use example images provided by user User specifies colors, textures, shapes User specifies image regions User specifies region relationships User sketches structures of images
Stockman MSU CSE3 Query by example User provides image (top left) System creates its own feature rep. to match to other images
Stockman MSU CSE4 QBIC (IBM) color histogram matching; user chooses colors
Stockman MSU CSE5 Query is grid painted by user
Stockman MSU CSE6 Texture features also possible
Stockman MSU CSE7 User can sketch objects (more research needed) User sketches boundaries of interest System will use elastic matching (see Ch 14 S&S) on images in DB Can be expensive
Stockman MSU CSE8 Results of elastic matching
Stockman MSU CSE9 Current problems Indexing needed for fast browsing, but how can indexes be built? Computing image features online will be slow, even offline computing will be slow. What about deeper queries: “show me pictures of children enjoying eating” (same problem faced by traditional databases) Show me pictures of tragic events, of poverty, of natural beauty, of triumph against bad odds …