Representing scenes for real-time context classification on mobile devices

Daniele Ravi

Abstract: This paper introduces the DCT-GIST image representation model for real-time scene context classification on mobile devices with limited memory and computational resources. The proposed approach exploits DCT coefficients statistics modeled as Laplacian distributions, which can be computed directly in the JPEG compressed domain without additional decoding. The method is coupled with Support Vector Machine classification and achieves competitive results on scene classification benchmarks with minimal computational overhead compared to traditional GIST descriptors.