Person Identification Using Discriminative Visual Aesthetic

A person’s image aesthetic is defined as a set of principles that influences the person to choose favorite images over a list of options. Different persons have different visual preferences which can be used to discriminate a person from another. Recently, some research has been carried in the area of behavioral biometric and image aesthetic. Researchers prove that it is possible to identify a person from discriminative visual cues. In this paper, we develop a new and improved method for person recognition using aesthetic features. The proposed approach uses 14 perceptual and 3 content features collected from the state-of-the-art researches. To achieve significant improvement in rank 1 recognition rate, we utilize local perceptual features and Histogram Oriented Gradient (HOG) feature for the first time. However, the new feature space is 975 dimensional which increases the elapsed time of enrollment and recognition phases. To minimize it, we apply Principle Component Analysis (PCA) that reduces the dimension of the feature vector by 50% without affecting the actual recognition performance. The proposed method has been evaluated on 200 user’s 40,000 images from the benchmark Flickr database. Experiment shows that the proposed method achieves 84% and 97% recognition rates in rank 1 and 5, whereas most well-performed state-of-the-art method shows 73% and 92%, respectively.