Intra-dimensional feature diagnosticity in the Fuzzy Feature Contrast Model

Similarity assessment is a basic operation to query images in a large database. Based on fuzzy logic, Santini and Jain extended Tversky’s Feature Contrast Model (FCM) to measure image similarity, and developed the Fuzzy Feature Contrast Model (FFCM). In this paper, we analyze the distinction between FCM and FFCM in terms of feature representations, and point out that the intra-dimensional feature diagnosticity in the FCM has not been considered in the FFCM. Consequently, similarity measures of the FFCM are positively correlated with visual feature intensities. In order to depress the positive correlation and preserve the original idea of the FCM where possible, we propose an extension of the FFCM called the Diagnostic Fuzzy Feature Contrast Model (DFFCM). Both the feature diagnosticity and feature intensity are employed to measure the image similarity by the DFFCM. The simulated experimental results demonstrated that the impact of the feature intensity on similarity measures of the DFFCM was weaker than that of the FFCM. Experimental results based on synthetic and real-word image databases showed that the DFFCM outperformed the FFCM in terms of image similarity measures.