Face Image Retrieval using Contextual and HAAR Feature Extraction

Photos with people (e.g., family, friends, celebrities, etc.) are the major interest of users. Many applications including face verification use face image retrieval method. A search-based face annotation framework investigated for mining weakly labelled facial images that are freely available on the internet. A key component of such a search-based annotation paradigm is building a database of facial images with accurate labels. Since the facial images are often noisy, to obtain ground truth labels, they manually examine the facial images and remove the irrelevant facial images for each name. As the database do not contain images with different poses, expressions and angles that are labelled accurately. In the proposed system, we automatically detected human attributes that contain semantic cues of the face photos to improve content based face retrieval by constructing semantic code-words for efficient large-scale face retrieval. By leveraging human attributes in a scalable framework, we propose one of two methods named attribute-enhanced sparse coding and attribute embedded inverting indexing to improve the face retrieval in the offline and online stages. For the effectiveness of different attributes a set of Haar-like features are extracted and contextual features are used to improve the accuracy and also to give ranked images of the retrieved image.