Sift Descriptors Modeling and Application in Texture Image Classification

Abstract

This paper presents a new statistical model for describing real textured images. Our model is based on the observation that the Scale-Invariant Feature Transform (SIFT) descriptors extracted from a given image can be properly modeled by the Gamma distribution. The maximum-likelihood algorithm was used to estimate the two parameters of the Gamma distribution. The efficiency of the proposed approach was validated in the classification stage. Experiments were conducted on Brodatz database. Results demonstrated that our model leads to good improvement in term of the accuracy rate.