Video Scene Analysis in 3D Wavelet Transform Domain

Zhi Li,Guizhong Liu

DOI: https://doi.org/10.1007/s11042-010-0594-z

IF: 2.577

2010-01-01

Multimedia Tools and Applications

Abstract:This paper proposes a novel scene analysis algorithm based on three-dimensional discrete wavelet transform (3D DWT). Based on the correlation among the adjacent frames, video frames can be considered as four categories: abrupt scene transition, motion scene, gradual scene transition and static scene, which are ranked from low to high according to the strength of the correlation. Through the investigation of the particular temporal and spatial distribution of each category, the correlation among adjacent frames could be described by the 3D DWT coefficients related statistical features, which are the energy of high-frequency coefficients difference, the sum of high-frequency coefficients magnitudes and the difference of low-frequency coefficients magnitudes. The energy of high-frequency coefficients difference is first used to detect the abrupt scene transition including cut and flashlight. Then all the three features are input to SVM for the purpose of analyzing the residual scenes and detecting the gradual scene transition, such as dissolve and fade. Experimental results show the method to be effective not only for the abrupt scene transition detection, but also for the gradual scene transition detection.