Semantic-Based Graph Index for Mobile Photo Search

Abstract

Nowadays, more and more photos are created by mobile devices equipped with camera. How to fast index and retrieve the photos is one key issue in photo management. In this paper, we present a semantic-based graph index for finding the relevant photos from one photo. First, we cluster photos into many different collections by single-dimension context like time, location, author and tag and form initial single-context photo clustering graphs. Second, we analyze the semantic relation of the context by semantic correlation to form single-context semantic relation graph. Third, we fuse the initial single-context photo clustering relation graph and single-context semantic relation graph into single-context photo semantic relation graph. Lastly, we combine the all single-context photo semantic relation graphs into a multi-dimensional photo semantic relation graph as indexing graph. The graph index method has been implemented in Kaleido Photo which is a photo-centric management system and test results show the method is efficient for photo indexing and retrieval.