Compressed domain zoom motion detection and classification based on application of local ternary patterns on block motion vectors

Abstract

This paper presents an ovel application of the local ternary patterns to the zoom motion detection problem and its further classification into zoom-in and zoom-out for compressed domain video sequences. The premise of the proposed method is based on modeling the compressed domain motion vector orientation information using local ternary patterns with a fixed neighborhood. The obtained ternary code is analyzed by forming the upper and lower patterns followed by converting these patterns into local binary patterns which is utilized for training the C-SVM classifier. Experimental testing using Exhaustive Search Motion Estimation obtained block motion vectors as well as H.264 obtained block motion vectors along with comparative analysis carried out with existing techniques shows superior performance for the proposed method.