Finding semantically related images in the WWW

Abstract

To represent the image semantics more adequately, we propose the Weight ChainNet model that is based on the concept of lexical chain. A lexical chain (LC) is a sequence of semantically related words in a text. Here, we define it as one sentence that carries certain semantics by its words. As an image title is just a single word, we say it’s a trivial lexical chain Title Lexical Chain (TLC). The text obtained from the ALT tag is referred to as the Alt Lexical Chain (ALC). The page title is represented as a LC too Page Lexical Chain (PLC). Finally, since a caption comprises multiple sentences, we represent it as three types of lexical chains. Type one is called sentence lexical chain (SLC), which represents one single sentence in an image caption. Type two is called reconstructed sentence lexical chain (RSLC), and it represents one new sentence reconstructed from related sentences. Two sentences are related if both share one or more words. One common word in two SLCs splits each SLC into two. Based on the first common word, the second SLC’s second half is connected to the first SLC’s first half to form a RSLC. The last type is called caption lexical chain (CLC), which represents the whole image caption. A CLC is formed by connecting SLC one after another.