Reversible Data Hiding Research On Infrared Image For Electrical Equipment

Fault detection and diagnosis technology based on thermal infrared image is an important solution for power equipment state overhaul.On-site information can be embedded into it in invisible way and extracted out losslessly by means of reversible data hiding method.On account of identifying image producing,this information can be used to realize the origin certification of huge amounts of thermal infrared images of electric equipments.This thesis focuses on the key technology in reversible data hiding with the research object of the field electrical equipmen images on working condition.The main research work is as follows.Usually,there are a lot of salt and pepper noises in electrical equipment thermal infrared images and they locate in random distritution.This leads to too long compressed singular pixel location map roomming for a large number of data.In order to solve this question,an adjustment method for high density jump location map is proposed.Short zero groups are modified in a strategy of discarding some extensible prediction errors.Only those under a certain threshold value get to be preserved.This method improves the continuity of the locations of the same attribute tags,compresses a great deal of drawing of redundancy and increases the net charge embedding capacity.Aiming at exploding the capacity restriction caused by single zero prediction error in pixel-based prediction algorithm;an integrated prediction-error extension method is put forward.The novel method referres the pixel prediction and error expansion in fixed block-based scanning.Two directions with independent extendding standard are designed.More pixels get prediction values for carring data bits.When the context pixel number is large,noise level is unable to describe the detailed characteristics in prediction block of the current pixel.An union pixel selection method is presented using extreme value depth and the local noise level.Multiple histogram modification is utilized to select high smoothness pixels to be processed for embedding data as a matter of priority.Taking advantage of match of prediction-error extension and noise level,peak signal-to-noise ratio is improved as a result.For single layer information embeddding,pixels are modified in single direction.A novel double-predictor and double-layer reversibledata hiding scheme is invented on the basis of partially offset.New method qualifies the second layer the ability to modify parts of the pixels in the opposite direction in the first layer embedding.After the whole embedding procedure,those pixels are restored to the status of information hiding before.Thus,a single pixel can carry two data bits at most.Consequently,payload capacity is enhanced and distortion leaded by this can also be decreased.