Statistics- and Spatiality-Based Feature Distance Measure for Error Resilient Image Authentication

Content-based image authentication typically assesses authenticity based on a distance measure between the image to be tested and its original. Commonly employed distance measures such as the Minkowski measures (including Hamming and Euclidean distances) may not be adequate for content-based image authentication since they do not exploit statistical and spatial properties in features. This paper proposes a feature distance measure for content-based image authentication based on statistical and spatial properties of the feature differences. The proposed statisticsand spatiality-based measure (SSM ) is motivated by an observation that most malicious manipulations are localized whereas acceptable manipulations result in global distortions. A statistical measure, kurtosis, is used to assess the shape of the feature difference distribution; a spatial measure, the maximum connected component size, is used to assess the degree of object concentration of the feature differences. The experimental results have confirmed that our proposed measure is better than previous measures in distinguishing malicious manipulations from acceptable ones.