Automatic Object Identification in an Annotated Image using Latent Semantic Analysis

Abstract

Object identification is the process of mapping visual areas within an image to human been concepts, this is a vital approach to unify image understanding between machines and human. Anyway, images in the Internet are retrieved using natural language query where search engines implement text match to the textual description attached to images. Current image retrieval engines produce huge disturbance in fulfilling user query due to the nature of text matching technique where many un-wanted results are obtained.
In this paper a new model is presented primarily to automatically identify objects composing an image and map identified objects to attached annotations; this is implemented in this paper by using LSA (Latent Semantic Analysis). This approach has lead to secondary revenue where annotations have been de-noised from an-wanted words. A real case study has been taken to verify the hypothesis of this paper and resultant calculations approved the approach.