The Detection Of Camera Sabotage Based On The Contourlet Transform

Nowadays surveillance cameras are widely used in all fields of social life, which makesgreat contributions to the harmony and peace of national life. However there are still somelawbreakers who do anything to sabotage and destroy the surveillance, such as covering thelens with the clothes, painting on the camera lens and move the camera to change the field ofthe designated view. Once the camera is sabotaged, the security agents are no longer able tomonitor the designated fields by the camera which leads to a loss of surveillance ability.Consequently it is the camera sabotage detection that becomes far higher on the agenda of theintelligent video surveillance.This paper firstly reviews the existing sabotage detection algorithms. Then based on theinsight of the pool-based camera tampering detection algorithm, we propose theimprovements:1. We substitute the Bhattacharyya Distance which has more accuracy andstatistical significance for the original histogram distance measure;2. We replace thehistogram measure based on Sobel kernel with the one based on the LOG kernel, which hasnot only the better anti-noise effect but also the more complete edge extraction. The algorithmproposed has more accuracy and robust in the experiments.Secondly we take an in-depth study of contourlet transform, the new generation of themulti-resolution and multi-directional decomposition. According to the study we finallypropose the sabotage detection algorithm based on contourlet transform. The contourlettransform is mainly comprised of two steps:1) Pyramid Frames, decompose the image into theband-pass and low frequency sub-images;2) Iterated Directional Filter Banks, decompose theband-pass image into different subband images of the same orientation. The detectionalgorithm based on contourlet transformation distinguishes the sabotaged events from othersby comparing the frames and the background frames on different image feature function,which largely depends on the contourlet coefficients.A large number of experiments show that our algorithm, compared to the existingalgorithm based on wavelet transform, has achieved better effect and presents a new application field for contourlet transformation.