Bradley Scott Denney

Scalable attribute-driven face image retrieval

In recent years an explosion of online multimedia data has been witnessed. As an example, abundant photos recording every aspect of human life are available through social media. Among tremendous amount of photos, a significant fraction contains human faces. Faces are usually salient features of the photos. To understand and extract useful information from such gigantic data corpus, efficient and effective retrieval algorithms are demanded. Most face retrieval techniques rely on low-level image features to compare faces based on visual similarity. However, as humans we tend to simplify the recognition task by utilizing human attributes such as gender or race to help differentiate people on a higher semantic level. In this paper, we propose to use human attributes as high-level semantic cues to determine peoples identities. To this end, we develop discriminative image features with attribute information encoded to achieve more accurate face image retrieval. To guarantee scalability, we propose using a binary coding scheme for the proposed attributed-based features. A re-ranking step after initial retrieval is incorporated to further improve the retrieval performance. We demonstrate the superiority of the proposed method compared to state-of-the-art on the LFW and Pubfig face datasets.

A collaborative approach for face verification and attributes refinement

Face verification which aims to determine whether two given faces refer to the same person, and human attribute learning with the goal of extracting predefined describable attributes from face images, are two fundamental issues in a variety of applications (e.g., face tagging, attribute based face search). While advances in computer vision domain have resulted in a series of techniques for each of the two tasks, such techniques are usually prone to errors due to the large variation of faces in pose, expression, illumination, occlusion, etc. Different from most prior related works which focus on the two tasks separately, in this paper, we explore their relationships and propose a collaborative approach allowing them to interact with each other to iteratively reduce errors and uncertainties. The interaction is embodies in two processes, one is that the results of face verification can be leveraged to refine attribute values utilizing the random-walk model, and the other is that the attribute values can also be employed to improve the face verification performance through R-LDA model. The two interactive processes will continue to iteratively improve the performance of the two tasks, until the relative stable results are achieved. Experimental results on the real-world photo collections demonstrate the effectiveness of the proposed approach.

Sub-event recognition and summarization for structured scenario photos

Structured scenario photos, referring to the images which capture important events that usually follow specific routines/structures (such as wedding ceremonies, graduation ceremonies, etc.), account for a significant proportion in personal photo collections. Conventional image analysis techniques without considering the event routines/structures are not sufficient to handle these photos. In this paper, we explore the appropriate framework to learn and utilize the specific routines for understanding these structure scenario photos. Specifically, we propose a novel framework which can systematically integrate Hidden Markov Model and Gaussian Mixture Model to recognize sub-events from structured scenario photos. Then we present a comprehensive criterion to select representative images to summarize the whole photo collection. Experimental results conducted on the real-world datasets demonstrate the superiority of our framework in both of sub-event recognition and photo summarization tasks.