Subspace mapping of facial attributes for face image retrieval
International Journal of Latest Trends in Engineering and Technology · 2017
Several research works had been endorsed in the last decade to develop image retrieval techniques from the multimedia databases.Although large number of indexing and retrieval techniques has been developed, there is no universally conventional feature extraction, indexing and retrieval technique obtainable.Among the vast digital images and photos shared on the internet, a big percentage of them are photos related to human faces.Many research problems and opportunities for a variety of real-world applications are based upon the face images.This has created an overriding need to provide proficient means of face image retrieval for many real-world applications.The designing aspect of efficient face image retrieval should take account of the issues such as noise, illumination, poses and expressions, low cost of processing and time management along with elimination of misclassified face images.The Extraction of low level components by normalising the appearance of face, selecting an suitable feature description and designing matching methodologies involves the challenge of incorporation of colour, texture and shape properties of the images.Accurate retrieval remains a difficult task for face images and is challenging due to the huge intra-class variation of same person, which is bigger than the inter-class variation between different persons.In previous works on face image retrieval, faces are typically described as low level image features and are sensitive to intra-class variations.Directly using high level attributes for retrieval, is also not precise since each individual attribute contains very limited information and is not quite enough to retrieve corresponding faces of a query subject.To improve the discriminative abilities of low level features and facial attributes, we propose a face image retrieval framework driven by enabled facial attributes.