Constructing a web image graph requires uses the VIPS algorithm  which separates out the web page into visual blocks that are organized in a tree. Different blocks are related to different topics, so we use the hyperlinks from block to page, rather than from page to page. The page to block and block to page relationships are extracted first. The block to page matrix  Z with dimensions nxk is constructed first  and the matrix has values for a cell as the inverse of the number of pages to which a block links or zero otherwise. The page to block matrix X with dimensions kxn is populated with a normalized importance value based on the size and the distance from the center or zero otherwise. These two matrix are combined to form a new web page graph W = ZX.  Let Y define a block to image matrix with dimension nxm such that each block contains the inverse of the number of images contained in the image block or zero otherwise. Using link level analysis, the block graph is defined as Wb = (1-t)ZX +TU/D where t is a constant and D is a diagonal matrix. The diagonal matrix is populated with zero if block i and block j are contained in two different web pages, otherwise it is set to the default degree of coherence from VIPS algorithm. The block graph attempts to define the probability of jumping from block a to block b. Then the image graph is constructed Wi = (Y-Transposed)WbY. This image graph better reflects semantic relationships between images and can be used with the data mining techniques discussed earlier.